Zhen Huang

Xiaoyuan Liu, Tianneng Shi, Chulin Xie et al.

Federated Learning (FL) has become a practical and widely adopted distributed learning paradigm. However, the lack of a comprehensive and standardized solution covering diverse use cases makes it challenging to use in practice. In addition, selecting an appropriate FL framework for a specific use case can be a daunting task. In this work, we present UniFed, the first unified platform for standardizing existing open-source FL frameworks. The platform streamlines the end-to-end workflow for distributed experimentation and deployment, encompassing 11 popular open-source FL frameworks. In particular, to address the substantial variations in workflows and data formats, UniFed introduces a configuration-based schema-enforced task specification, offering 20 editable fields. UniFed also provides functionalities such as distributed execution management, logging, and data analysis. With UniFed, we evaluate and compare 11 popular FL frameworks from the perspectives of functionality, privacy protection, and performance, through conducting developer surveys and code-level investigation. We collect 15 diverse FL scenario setups (e.g., horizontal and vertical settings) for FL framework evaluation. This comprehensive evaluation allows us to analyze both model and system performance, providing detailed comparisons and offering recommendations for framework selection. UniFed simplifies the process of selecting and utilizing the appropriate FL framework for specific use cases, while enabling standardized distributed experimentation and deployment. Our results and analysis based on experiments with up to 178 distributed nodes provide valuable system design and deployment insights, aiming to empower practitioners in their pursuit of effective FL solutions.

Wei Li, Qiming Zhang, Jing Zhang et al.

Single image deraining (SID) in real scenarios attracts increasing attention in recent years. Due to the difficulty in obtaining real-world rainy/clean image pairs, previous real datasets suffer from low-resolution images, homogeneous rain streaks, limited background variation, and even misalignment of image pairs, resulting in incomprehensive evaluation of SID methods. To address these issues, we establish a new high-quality dataset named RealRain-1k, consisting of $1,120$ high-resolution paired clean and rainy images with low- and high-density rain streaks, respectively. Images in RealRain-1k are automatically generated from a large number of real-world rainy video clips through a simple yet effective rain density-controllable filtering method, and have good properties of high image resolution, background diversity, rain streaks variety, and strict spatial alignment. RealRain-1k also provides abundant rain streak layers as a byproduct, enabling us to build a large-scale synthetic dataset named SynRain-13k by pasting the rain streak layers on abundant natural images. Based on them and existing datasets, we benchmark more than 10 representative SID methods on three tracks: (1) fully supervised learning on RealRain-1k, (2) domain generalization to real datasets, and (3) syn-to-real transfer learning. The experimental results (1) show the difference of representative methods in image restoration performance and model complexity, (2) validate the significance of the proposed datasets for model generalization, and (3) provide useful insights on the superiority of learning from diverse domains and shed lights on the future research on real-world SID. The datasets will be released at https://github.com/hiker-lw/RealRain-1k

Noah Amsel, Yves Baumann, Paul Beckman et al. · berkeley

This document presents a series of open questions arising in matrix computations, i.e., the numerical solution of linear algebra problems. It is a result of working groups at the workshop Linear Systems and Eigenvalue Problems, which was organized at the Simons Institute for the Theory of Computing program on Complexity and Linear Algebra in Fall 2025. The complexity and numerical solution of linear algebra problems is a crosscutting area between theoretical computer science and numerical analysis. The value of the particular problem formulations here is that they were produced via discussions between researchers from both groups. The open questions are organized in five categories: iterative solvers for linear systems, eigenvalue computation, low-rank approximation, randomized sketching, and other areas including tensors, quantum systems, and matrix functions.

Guanlong Wu, Ju Yang, Zhen Huang et al.

Trusted Execution Environments (TEEs)-aided federated learning protocols emerge as promising solutions to counter server-side adversaries and ensure the trustworthiness of the server. In this paper, we dissect existing protocols and demonstrate that server-side adversaries can still manipulate client selection and replay aggregation to compromise system robustness and privacy, by exploiting TEE limitations, i.e., state rollback and I/O manipulation. To this end, we present DIST-FL, a distributed system of servers guarded by multiple TEEs forming an append-only ledger for privacy-preserved, robust FL aggregation. Specifically, DIST-FL ensures operation linearizability to thwart state rollback attacks and incorporates inputs from reliable servers to mitigate I/O manipulation threats. We implement DIST-FL and conduct evaluations in WAN settings. Experimental results demonstrate that DIST-FL can effectively counter the proposed attacks and match the single-TEE's performance while offering a 6x throughput boost over its counterparts, leveraging TEE's computational advantages.

Wei Chen, Zhen Huang, Liang Xie et al.

Large Language Models (LLMs) tend to prioritize adherence to user prompts over providing veracious responses, leading to the sycophancy issue. When challenged by users, LLMs tend to admit mistakes and provide inaccurate responses even if they initially provided the correct answer. Recent works propose to employ supervised fine-tuning (SFT) to mitigate the sycophancy issue, while it typically leads to the degeneration of LLMs' general capability. To address the challenge, we propose a novel supervised pinpoint tuning (SPT), where the region-of-interest modules are tuned for a given objective. Specifically, SPT first reveals and verifies a small percentage (<5%) of the basic modules, which significantly affect a particular behavior of LLMs. i.e., sycophancy. Subsequently, SPT merely fine-tunes these identified modules while freezing the rest. To verify the effectiveness of the proposed SPT, we conduct comprehensive experiments, demonstrating that SPT significantly mitigates the sycophancy issue of LLMs (even better than SFT). Moreover, SPT introduces limited or even no side effects on the general capability of LLMs. Our results shed light on how to precisely, effectively, and efficiently explain and improve the targeted ability of LLMs. Code and data are available at https://github.com/yellowtownhz/sycophancy-interpretability.

Haowen Hou, Zhen Huang, Zheming Liang et al.

Video is temporally redundant: adjacent frames usually share most objects, background, and layout. Yet existing video multimodal large language models (video MLLMs) usually encode each sampled frame as an independent RGB image, causing visual tokens to repeat content already present in earlier frames. This suggests a more direct video interface: send a full reference frame only when the scene cannot be predicted well from prior context, and otherwise transmit a compact description of inter-frame changes. We call this interface a \emph{predictive visual code}, and instantiate it for video MLLMs as \textbf{AdaCodec}. AdaCodec spends full visual tokens on a reference frame only when its conditional predictive cost is high; otherwise, it encodes inter-frame changes, including motion and prediction residuals, as compact P-tokens. Across all eleven benchmarks, AdaCodec improves over the Qwen3-VL-8B per-frame RGB baseline at a matched visual-token budget. Even at $1/7$ the budget, AdaCodec with 32k tokens surpasses the 224k baseline on all long-video benchmarks; on five general-video benchmarks, it raises the average score while substantially cutting time-to-first-token from 9.26s to 1.62s.

Shitong Shao, Xiaohan Yuan, Zhen Huang et al.

Dataset expansion can effectively alleviate the problem of data scarcity for medical image segmentation, due to privacy concerns and labeling difficulties. However, existing expansion algorithms still face great challenges due to their inability of guaranteeing the diversity of synthesized images with paired segmentation masks. In recent years, Diffusion Probabilistic Models (DPMs) have shown powerful image synthesis performance, even better than Generative Adversarial Networks. Based on this insight, we propose an approach called DiffuseExpand for expanding datasets for 2D medical image segmentation using DPM, which first samples a variety of masks from Gaussian noise to ensure the diversity, and then synthesizes images to ensure the alignment of images and masks. After that, DiffuseExpand chooses high-quality samples to further enhance the effectiveness of data expansion. Our comparison and ablation experiments on COVID-19 and CGMH Pelvis datasets demonstrate the effectiveness of DiffuseExpand. Our code is released at https://github.com/shaoshitong/DiffuseExpand.

Hao Wang, Yangguang Li, Zhen Huang et al.

With the rapid development of automatic fake news detection technology, fact extraction and verification (FEVER) has been attracting more attention. The task aims to extract the most related fact evidences from millions of open-domain Wikipedia documents and then verify the credibility of corresponding claims. Although several strong models have been proposed for the task and they have made great progress, we argue that they fail to utilize multi-view contextual information and thus cannot obtain better performance. In this paper, we propose to integrate multi-view contextual information (IMCI) for fact extraction and verification. For each evidence sentence, we define two kinds of context, i.e. intra-document context and inter-document context}. Intra-document context consists of the document title and all the other sentences from the same document. Inter-document context consists of all other evidences which may come from different documents. Then we integrate the multi-view contextual information to encode the evidence sentences to handle the task. Our experimental results on FEVER 1.0 shared task show that our IMCI framework makes great progress on both fact extraction and verification, and achieves state-of-the-art performance with a winning FEVER score of 72.97% and label accuracy of 75.84% on the online blind test set. We also conduct ablation study to detect the impact of multi-view contextual information. Our codes will be released at https://github.com/phoenixsecularbird/IMCI.

Hao Wang, Yixin Cao, Yangguang Li et al.

Document-level natural language inference (DOCNLI) is a new challenging task in natural language processing, aiming at judging the entailment relationship between a pair of hypothesis and premise documents. Current datasets and baselines largely follow sentence-level settings, but fail to address the issues raised by longer documents. In this paper, we establish a general solution, named Retrieval, Reading and Fusion (R2F) framework, and a new setting, by analyzing the main challenges of DOCNLI: interpretability, long-range dependency, and cross-sentence inference. The basic idea of the framework is to simplify document-level task into a set of sentence-level tasks, and improve both performance and interpretability with the power of evidence. For each hypothesis sentence, the framework retrieves evidence sentences from the premise, and reads to estimate its credibility. Then the sentence-level results are fused to judge the relationship between the documents. For the setting, we contribute complementary evidence and entailment label annotation on hypothesis sentences, for interpretability study. Our experimental results show that R2F framework can obtain state-of-the-art performance and is robust for diverse evidence retrieval methods. Moreover, it can give more interpretable prediction results. Our model and code are released at https://github.com/phoenixsecularbird/R2F.

Yangbo Wei, Zhen Huang, Shaoqiang Lu et al.

Recent self-evolving agents have shown that skills can be discovered, refined, and accumulated through execution. However, existing skill-evolution frameworks typically assume a fixed tool layer and evaluate each skill independently, limiting their ability to repair tool-level failures or reason about interactions among skills. We propose SkillSmith, a synergy-aware skill-tool co-evolution framework. SkillSmith introduces a unified proposal space in which reflection produces atomic bundles that jointly modify skills and tools, allowing tools to be wrapped, edited, composed, split, or retired when skill evolution identifies a reusable capability gap. To guide this joint search, SkillSmith maintains an ecological utility model inspired by Lotka-Volterra dynamics, where an interaction matrix estimated from execution traces captures pairwise complementarity and conflict among skills and provides pressure signals for retrieval, mutation prioritization, and retirement. Furthermore, SkillSmith records anti-patterns, including failure signatures, causal attributions, and remedies, to accelerate diagnosis and veto proposals that repeat known mistakes. Experiments on three benchmarks, including WildClawBench, and five Qwen3.5 model scales show that SkillSmith consistently outperforms strong baselines, with gains that amplify as task complexity and multi-skill co-activation increase.

Zhen Huang, Haiyang Xin, Wenkai Yang et al.

Data-driven thermal predictors for 3D-ICs are often trained from scratch for each chip design using many high-fidelity finite-element simulations, leading to high data-generation cost and costly cross-design reuse. We propose Therm-FM, a neural operator framework that adapts a pretrained partial differential equation (PDE) foundation model to steady-state and transient 3D-IC thermal simulation. The motivation is that steady-state and transient chip-level heat conduction respectively share elliptic and parabolic operator structures with diffusion-type PDEs, allowing pretrained diffusion priors to provide an effective initialization for thermal-field prediction under heterogeneous materials, dense TSV/microbump interconnects, and package-level boundary conditions. To further reduce data-generation cost, Therm-FM incorporates a thermal-equivalent multi-fidelity training strategy that uses low-cost approximate simulations for thermal-domain adaptation and limited high-fidelity samples for calibration. Experiments on public HotSpot benchmarks and industrial 3D-IC package benchmarks show that Therm-FM achieves up to a 10.6x reduction in mean error and surpasses prior best accuracy with less than 20% of the training data. In cross-chip adaptation, it matches or surpasses full-data baselines in several metrics using only 10--30 target samples. We release datasets, source code, and pretrained models at https://github.com/haiyangxin/Therm-FM.

DeepSeek-AI, Daya Guo, Dejian Yang et al. · stanford, tsinghua

We introduce our first-generation reasoning models, DeepSeek-R1-Zero and DeepSeek-R1. DeepSeek-R1-Zero, a model trained via large-scale reinforcement learning (RL) without supervised fine-tuning (SFT) as a preliminary step, demonstrates remarkable reasoning capabilities. Through RL, DeepSeek-R1-Zero naturally emerges with numerous powerful and intriguing reasoning behaviors. However, it encounters challenges such as poor readability, and language mixing. To address these issues and further enhance reasoning performance, we introduce DeepSeek-R1, which incorporates multi-stage training and cold-start data before RL. DeepSeek-R1 achieves performance comparable to OpenAI-o1-1217 on reasoning tasks. To support the research community, we open-source DeepSeek-R1-Zero, DeepSeek-R1, and six dense models (1.5B, 7B, 8B, 14B, 32B, 70B) distilled from DeepSeek-R1 based on Qwen and Llama.

DeepSeek-AI, Aixin Liu, Bei Feng et al. · pku

We present DeepSeek-V2, a strong Mixture-of-Experts (MoE) language model characterized by economical training and efficient inference. It comprises 236B total parameters, of which 21B are activated for each token, and supports a context length of 128K tokens. DeepSeek-V2 adopts innovative architectures including Multi-head Latent Attention (MLA) and DeepSeekMoE. MLA guarantees efficient inference through significantly compressing the Key-Value (KV) cache into a latent vector, while DeepSeekMoE enables training strong models at an economical cost through sparse computation. Compared with DeepSeek 67B, DeepSeek-V2 achieves significantly stronger performance, and meanwhile saves 42.5% of training costs, reduces the KV cache by 93.3%, and boosts the maximum generation throughput to 5.76 times. We pretrain DeepSeek-V2 on a high-quality and multi-source corpus consisting of 8.1T tokens, and further perform Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL) to fully unlock its potential. Evaluation results show that, even with only 21B activated parameters, DeepSeek-V2 and its chat versions still achieve top-tier performance among open-source models.

Zhen Huang, Zhihuang Liu, Weishang Wu et al.

Large language models (LLMs) are increasingly used as general planners in embodied intelligence, enabling high level coordination and low level task planning for both single robot and multi-robot collaboration. This increasing reliance on embodied LLM planners also raises critical security concerns, since misaligned or manipulated instructions can be translated into physical actions. Prior work has studied such threats in single robot settings, while security risks in LLM controlled multi-robot collaboration, especially those propagated through inter robot communication, remain largely unexplored. To bridge this gap, we propose a novel attack paradigm for multi-robot system in which the adversary interacts with only a single entry robot. The compromised robot then propagates malicious intent through peer communication, leading to coordinated unsafe actions across the system. Our evaluation, covering high risk dimensions of dereliction of duty, privacy compromise, and public safety hazards, reveals a persistent safety alignment gap in multi-robot planners. We quantify this process with three metrics, obedience, infectiousness, and stealthiness. Experiments demonstrate both persistent attacker control and rapid propagation: obedience reaches 1.00 in the strongest cases, and infectiousness rises to 0.90. Notably, the attack is highly efficient, requiring as few as 3.0 rounds to compromise all the robots while maintaining a stealthiness score of 0.81. Such risks are amplified when robots must resolve trade offs in critical situations, such as emergencies or conflicts of rights, because the coordination mechanism can unintentionally allow adversarial instructions to override safety requirements. The code is available at https://github.com/TheFatInsect/InfectBot.

Yunjian Jia, Zhen Huang, Jiping Yan et al.

Deploying federated learning at the wireless edge introduces federated edge learning (FEEL). Given FEEL's limited communication resources and potential mislabeled data on devices, improper resource allocation or data selection can hurt convergence speed and increase training costs. Thus, to realize an efficient FEEL system, this paper emphasizes jointly optimizing resource allocation and data selection. Specifically, in this work, through rigorously modeling the training process and deriving an upper bound on FEEL's one-round convergence rate, we establish a problem of joint resource allocation and data selection, which, unfortunately, cannot be solved directly. Toward this end, we equivalently transform the original problem into a solvable form via a variable substitution and then break it into two subproblems, that is, the resource allocation problem and the data selection problem. The two subproblems are mixed-integer non-convex and integer non-convex problems, respectively, and achieving their optimal solutions is a challenging task. Based on the matching theory and applying the convex-concave procedure and gradient projection methods, we devise a low-complexity suboptimal algorithm for the two subproblems, respectively. Finally, the superiority of our proposed scheme of joint resource allocation and data selection is validated by numerical results.

Zhongwu Chen, Chengjin Xu, Fenglong Su et al.

Temporal relation prediction in incomplete temporal knowledge graphs (TKGs) is a popular temporal knowledge graph completion (TKGC) problem in both transductive and inductive settings. Traditional embedding-based TKGC models (TKGE) rely on structured connections and can only handle a fixed set of entities, i.e., the transductive setting. In the inductive setting where test TKGs contain emerging entities, the latest methods are based on symbolic rules or pre-trained language models (PLMs). However, they suffer from being inflexible and not time-specific, respectively. In this work, we extend the fully-inductive setting, where entities in the training and test sets are totally disjoint, into TKGs and take a further step towards a more flexible and time-sensitive temporal relation prediction approach SST-BERT, incorporating Structured Sentences with Time-enhanced BERT. Our model can obtain the entity history and implicitly learn rules in the semantic space by encoding structured sentences, solving the problem of inflexibility. We propose to use a time masking MLM task to pre-train BERT in a corpus rich in temporal tokens specially generated for TKGs, enhancing the time sensitivity of SST-BERT. To compute the probability of occurrence of a target quadruple, we aggregate all its structured sentences from both temporal and semantic perspectives into a score. Experiments on the transductive datasets and newly generated fully-inductive benchmarks show that SST-BERT successfully improves over state-of-the-art baselines.

Zhongwu Chen, Chengjin Xu, Fenglong Su et al.

In the last few years, the solution to Knowledge Graph (KG) completion via learning embeddings of entities and relations has attracted a surge of interest. Temporal KGs(TKGs) extend traditional Knowledge Graphs (KGs) by associating static triples with timestamps forming quadruples. Different from KGs and TKGs in the transductive setting, constantly emerging entities and relations in incomplete TKGs create demand to predict missing facts with unseen components, which is the extrapolation setting. Traditional temporal knowledge graph embedding (TKGE) methods are limited in the extrapolation setting since they are trained within a fixed set of components. In this paper, we propose a Meta-Learning based Temporal Knowledge Graph Extrapolation (MTKGE) model, which is trained on link prediction tasks sampled from the existing TKGs and tested in the emerging TKGs with unseen entities and relations. Specifically, we meta-train a GNN framework that captures relative position patterns and temporal sequence patterns between relations. The learned embeddings of patterns can be transferred to embed unseen components. Experimental results on two different TKG extrapolation datasets show that MTKGE consistently outperforms both the existing state-of-the-art models for knowledge graph extrapolation and specifically adapted KGE and TKGE baselines.

Shitong Shao, Huanran Chen, Zhen Huang et al.

In real teaching scenarios, an excellent teacher always teaches what he (or she) is good at but the student is not. This gives the student the best assistance in making up for his (or her) weaknesses and becoming a good one overall. Enlightened by this, we introduce the "Teaching what you Should Teach" strategy into a knowledge distillation framework, and propose a data-based distillation method named "TST" that searches for desirable augmented samples to assist in distilling more efficiently and rationally. To be specific, we design a neural network-based data augmentation module with priori bias, which assists in finding what meets the teacher's strengths but the student's weaknesses, by learning magnitudes and probabilities to generate suitable data samples. By training the data augmentation module and the generalized distillation paradigm in turn, a student model is learned with excellent generalization ability. To verify the effectiveness of our method, we conducted extensive comparative experiments on object recognition, detection, and segmentation tasks. The results on the CIFAR-10, ImageNet-1k, MS-COCO, and Cityscapes datasets demonstrate that our method achieves state-of-the-art performance on almost all teacher-student pairs. Furthermore, we conduct visualization studies to explore what magnitudes and probabilities are needed for the distillation process.

Menglong Lu, Zhen Huang, Yunxiang Zhao et al.

Self-training emerges as an important research line on domain adaptation. By taking the model's prediction as the pseudo labels of the unlabeled data, self-training bootstraps the model with pseudo instances in the target domain. However, the prediction errors of pseudo labels (label noise) challenge the performance of self-training. To address this problem, previous approaches only use reliable pseudo instances, i.e., pseudo instances with high prediction confidence, to retrain the model. Although these strategies effectively reduce the label noise, they are prone to miss the hard examples. In this paper, we propose a new self-training framework for domain adaptation, namely Domain adversarial learning enhanced Self-Training Framework (DaMSTF). Firstly, DaMSTF involves meta-learning to estimate the importance of each pseudo instance, so as to simultaneously reduce the label noise and preserve hard examples. Secondly, we design a meta constructor for constructing the meta-validation set, which guarantees the effectiveness of the meta-learning module by improving the quality of the meta-validation set. Thirdly, we find that the meta-learning module suffers from the training guidance vanishment and tends to converge to an inferior optimal. To this end, we employ domain adversarial learning as a heuristic neural network initialization method, which can help the meta-learning module converge to a better optimal. Theoretically and experimentally, we demonstrate the effectiveness of the proposed DaMSTF. On the cross-domain sentiment classification task, DaMSTF improves the performance of BERT with an average of nearly 4%.

Zhihong Lei, Xingyu Na, Mingbin Xu et al.

Large language models (LLMs) have shown superb capability of modeling multimodal signals including audio and text, allowing the model to generate spoken or textual response given a speech input. However, it remains a challenge for the model to recognize personal named entities, such as contacts in a phone book, when the input modality is speech. In this work, we start with a speech recognition task and propose a retrieval-based solution to contextualize the LLM: we first let the LLM detect named entities in speech without any context, then use this named entity as a query to retrieve phonetically similar named entities from a personal database and feed them to the LLM, and finally run context-aware LLM decoding. In a voice assistant task, our solution achieved up to 30.2% relative word error rate reduction and 73.6% relative named entity error rate reduction compared to a baseline system without contextualization. Notably, our solution by design avoids prompting the LLM with the full named entity database, making it highly efficient and applicable to large named entity databases.

Menglong Lu, Zhen Huang, Zhiliang Tian et al.

Text classification is a fundamental task for natural language processing, and adapting text classification models across domains has broad applications. Self-training generates pseudo-examples from the model's predictions and iteratively trains on the pseudo-examples, i.e., minimizes the loss on the source domain and the Gibbs entropy on the target domain. However, Gibbs entropy is sensitive to prediction errors, and thus, self-training tends to fail when the domain shift is large. In this paper, we propose Meta-Tsallis Entropy minimization (MTEM), which applies a meta-learning algorithm to optimize the instance adaptive Tsallis entropy on the target domain. To reduce the computation cost of MTEM, we propose an approximation technique to approximate the Second-order derivation involved in the meta-learning. To efficiently generate pseudo labels, we propose an annealing sampling mechanism for exploring the model's prediction probability. Theoretically, we prove the convergence of the meta-learning algorithm in MTEM and analyze the effectiveness of MTEM in achieving domain adaptation. Experimentally, MTEM improves the adaptation performance of BERT with an average of 4 percent on the benchmark dataset.

Xinyue Fang, Zhen Huang, Zhiliang Tian et al.

LLMs obtain remarkable performance but suffer from hallucinations. Most research on detecting hallucination focuses on the questions with short and concrete correct answers that are easy to check the faithfulness. Hallucination detections for text generation with open-ended answers are more challenging. Some researchers use external knowledge to detect hallucinations in generated texts, but external resources for specific scenarios are hard to access. Recent studies on detecting hallucinations in long text without external resources conduct consistency comparison among multiple sampled outputs. To handle long texts, researchers split long texts into multiple facts and individually compare the consistency of each pairs of facts. However, these methods (1) hardly achieve alignment among multiple facts; (2) overlook dependencies between multiple contextual facts. In this paper, we propose a graph-based context-aware (GCA) hallucination detection for text generations, which aligns knowledge facts and considers the dependencies between contextual knowledge triples in consistency comparison. Particularly, to align multiple facts, we conduct a triple-oriented response segmentation to extract multiple knowledge triples. To model dependencies among contextual knowledge triple (facts), we construct contextual triple into a graph and enhance triples' interactions via message passing and aggregating via RGCN. To avoid the omission of knowledge triples in long text, we conduct a LLM-based reverse verification via reconstructing the knowledge triples. Experiments show that our model enhances hallucination detection and excels all baselines.

Zhihua Wen, Zhiliang Tian, Wei Wu et al.

Conditional story generation is significant in human-machine interaction, particularly in producing stories with complex plots. While Large language models (LLMs) perform well on multiple NLP tasks, including story generation, it is challenging to generate stories with both complex and creative plots. Existing methods often rely on detailed prompts to guide LLMs to meet target conditions, which inadvertently restrict the creative potential of the generated stories. We argue that leveraging information from exemplary human-written stories facilitates generating more diverse plotlines. Delving deeper into story details helps build complex and credible plots. In this paper, we propose a retrieval-au\textbf{G}mented sto\textbf{R}y generation framework with a f\textbf{O}rest of e\textbf{V}id\textbf{E}nce (GROVE) to enhance stories' complexity. We build a retrieval repository for target conditions to produce few-shot examples to prompt LLMs. Additionally, we design an ``asking-why'' prompting scheme that extracts a forest of evidence, providing compensation for the ambiguities that may occur in the generated story. This iterative process uncovers underlying story backgrounds. Finally, we select the most fitting chains of evidence from the evidence forest and integrate them into the generated story, thereby enhancing the narrative's complexity and credibility. Experimental results and numerous examples verify the effectiveness of our method.

Zhihong Lei, Ernest Pusateri, Shiyi Han et al.

Recent advances in deep learning and automatic speech recognition have improved the accuracy of end-to-end speech recognition systems, but recognition of personal content such as contact names remains a challenge. In this work, we describe our personalization solution for an end-to-end speech recognition system based on connectionist temporal classification. Building on previous work, we present a novel method for generating additional subword tokenizations for personal entities from their pronunciations. We show that using this technique in combination with two established techniques, contextual biasing and wordpiece prior normalization, we are able to achieve personal named entity accuracy on par with a competitive hybrid system.

Yiyuan Li, Zhen Huang, Yanan Wu et al.

The reasoning ability of large language models (LLMs) can be unleashed with reinforcement learning (RL) (OpenAI, 2024; DeepSeek-AI et al., 2025a; Zeng et al., 2025). The success of existing RL attempts in LLMs usually rely on high-quality samples of large volumes. In this paper, we challenge conventional assumptions about data requirements in RL for LLMs by demonstrating the effectiveness of one-shot reinforcement learning. Specifically, we introduce polymath learning, a framework for designing one training sample that elicits multidisciplinary reasoning improvement. We present three key findings: (1) A single, strategically selected math reasoning sample can produce significant performance improvements across multiple domains, including physics, chemistry, and biology; (2) Analysis of salient mathematical skills provides insight into the characteristics associated with effective polymath samples; and (3) An engineered synthetic sample that integrates multidisciplinary elements and broader skill coverage achieves stronger performance than naturally occurring individual samples. Across various reasoning benchmarks, polymath learning achieves stronger performance than larger datasets, demonstrating that reasoning structure and skills in samples, rather than quantity, may be the key to unlock enhanced reasoning capabilities in language models. Our results suggest a shift, dubbed as sample engineering, toward precision engineering of samples that complements simply increasing data volume.

Zhihua Wen, Zhiliang Tian, Zexin Jian et al.

Large Language Models (LLMs) are widely used for knowledge-seeking yet suffer from hallucinations. The knowledge boundary (KB) of an LLM limits its factual understanding, beyond which it may begin to hallucinate. Investigating the perception of LLMs' KB is crucial for detecting hallucinations and LLMs' reliable generation. Current studies perceive LLMs' KB on questions with a concrete answer (close-ended questions) while paying limited attention to semi-open-ended questions (SoeQ) that correspond to many potential answers. Some researchers achieve it by judging whether the question is answerable or not. However, this paradigm is unsuitable for SoeQ, which are usually partially answerable, containing both answerable and ambiguous (unanswerable) answers. Ambiguous answers are essential for knowledge-seeking, but they may go beyond the KB of LLMs. In this paper, we perceive the LLMs' KB with SoeQ by discovering more ambiguous answers. First, we apply an LLM-based approach to construct SoeQ and obtain answers from a target LLM. Unfortunately, the output probabilities of mainstream black-box LLMs are inaccessible to sample for low-probability ambiguous answers. Therefore, we apply an open-sourced auxiliary model to explore ambiguous answers for the target LLM. We calculate the nearest semantic representation for existing answers to estimate their probabilities, with which we reduce the generation probability of high-probability answers to achieve a more effective generation. Finally, we compare the results from the RAG-based evaluation and LLM self-evaluation to categorize four types of ambiguous answers that are beyond the KB of the target LLM. Following our method, we construct a dataset to perceive the KB for GPT-4. We find that GPT-4 performs poorly on SoeQ and is often unaware of its KB. Besides, our auxiliary model, LLaMA-2-13B, is effective in discovering more ambiguous answers.

DeepSeek-AI, Aixin Liu, Aoxue Mei et al.

We introduce DeepSeek-V3.2, a model that harmonizes high computational efficiency with superior reasoning and agent performance. The key technical breakthroughs of DeepSeek-V3.2 are as follows: (1) DeepSeek Sparse Attention (DSA): We introduce DSA, an efficient attention mechanism that substantially reduces computational complexity while preserving model performance in long-context scenarios. (2) Scalable Reinforcement Learning Framework: By implementing a robust reinforcement learning protocol and scaling post-training compute, DeepSeek-V3.2 performs comparably to GPT-5. Notably, our high-compute variant, DeepSeek-V3.2-Speciale, surpasses GPT-5 and exhibits reasoning proficiency on par with Gemini-3.0-Pro, achieving gold-medal performance in both the 2025 International Mathematical Olympiad (IMO) and the International Olympiad in Informatics (IOI). (3) Large-Scale Agentic Task Synthesis Pipeline: To integrate reasoning into tool-use scenarios, we developed a novel synthesis pipeline that systematically generates training data at scale. This methodology facilitates scalable agentic post-training, yielding substantial improvements in generalization and instruction-following robustness within complex, interactive environments.

Zhihong Lei, Mingbin Xu, Shiyi Han et al.

Recent advances in deep learning and automatic speech recognition (ASR) have enabled the end-to-end (E2E) ASR system and boosted the accuracy to a new level. The E2E systems implicitly model all conventional ASR components, such as the acoustic model (AM) and the language model (LM), in a single network trained on audio-text pairs. Despite this simpler system architecture, fusing a separate LM, trained exclusively on text corpora, into the E2E system has proven to be beneficial. However, the application of LM fusion presents certain drawbacks, such as its inability to address the domain mismatch issue inherent to the internal AM. Drawing inspiration from the concept of LM fusion, we propose the integration of an external AM into the E2E system to better address the domain mismatch. By implementing this novel approach, we have achieved a significant reduction in the word error rate, with an impressive drop of up to 14.3% across varied test sets. We also discovered that this AM fusion approach is particularly beneficial in enhancing named entity recognition.

Ji Zeng, Dayuan Fu, Tiantian Mi et al.

Recently, the frontier of Large Language Model (LLM) capabilities has shifted from single-turn code generation to agentic software engineering-a paradigm where models autonomously navigate, edit, and test complex repositories. While post-training methods have become the de facto approach for code agents, **agentic mid-training**-mid-training (MT) on large-scale data that mirrors authentic agentic workflows-remains critically underexplored due to substantial resource requirements, despite offering a more scalable path to instilling foundational agentic behaviors than relying solely on expensive reinforcement learning. A central challenge in realizing effective agentic mid-training is the distribution mismatch between static training data and the dynamic, feedback-rich environment of real development. To address this, we present a systematic study of agentic mid-training, establishing both the data synthesis principles and training methodology for effective agent development at scale. Central to our approach is **agent-native data**-supervision comprising two complementary types of trajectories: **contextually-native trajectories** that preserve the complete information flow an agent experiences, offering broad coverage and diversity; and **environmentally-native trajectories** collected from executable repositories where observations stem from actual tool invocations and test executions, providing depth and interaction authenticity. We verify the model's agentic capabilities on `SWE-Bench Verified`. We demonstrate our superiority over the previous open software engineering mid-training recipe `Kimi-Dev` under two post-training settings with an aligned base model and agentic scaffold, while using less than half mid-training tokens (73.1B). Besides relative advantage, our best performing 32B and 72B models achieve **56.1%** and **58.5%** resolution rates, respectively, which are ...

Ruoyu Chen, Siyuan Liang, Jingzhi Li et al.

Advances in multimodal pre-training have propelled object-level foundation models, such as Grounding DINO and Florence-2, in tasks like visual grounding and object detection. However, interpreting these models' decisions has grown increasingly challenging. Existing interpretable attribution methods for object-level task interpretation have notable limitations: (1) gradient-based methods lack precise localization due to visual-textual fusion in foundation models, and (2) perturbation-based methods produce noisy saliency maps, limiting fine-grained interpretability. To address these, we propose a Visual Precision Search method that generates accurate attribution maps with fewer regions. Our method bypasses internal model parameters to overcome attribution issues from multimodal fusion, dividing inputs into sparse sub-regions and using consistency and collaboration scores to accurately identify critical decision-making regions. We also conducted a theoretical analysis of the boundary guarantees and scope of applicability of our method. Experiments on RefCOCO, MS COCO, and LVIS show our approach enhances object-level task interpretability over SOTA for Grounding DINO and Florence-2 across various evaluation metrics, with faithfulness gains of 23.7%, 31.6%, and 20.1% on MS COCO, LVIS, and RefCOCO for Grounding DINO, and 102.9% and 66.9% on MS COCO and RefCOCO for Florence-2. Additionally, our method can interpret failures in visual grounding and object detection tasks, surpassing existing methods across multiple evaluation metrics. The code will be released at https://github.com/RuoyuChen10/VPS.

Adnan Haider, Tim Ng, Zhen Huang et al.

Maximum mutual information (MMI) has become one of the two de facto methods for sequence-level training of speech recognition acoustic models. This paper aims to isolate, identify and bring forward the implicit modelling decisions induced by the design implementation of standard finite state transducer (FST) lattice based MMI training framework. The paper particularly investigates the necessity to maintain a preselected numerator alignment and raises the importance of determinizing FST denominator lattices on the fly. The efficacy of employing on the fly FST lattice determinization is mathematically shown to guarantee discrimination at the hypothesis level and is empirically shown through training deep CNN models on a 18K hours Mandarin dataset and on a 2.8K hours English dataset. On assistant and dictation tasks, the approach achieves between 2.3-4.6% relative WER reduction (WERR) over the standard FST lattice based approach.

Zhen Huang, Yihao Li, Dong Pei et al.

Text-driven fashion synthesis and design is an extremely valuable part of artificial intelligence generative content(AIGC), which has the potential to propel a tremendous revolution in the traditional fashion industry. To advance the research on text-driven fashion synthesis and design, we introduce a new dataset comprising a million high-resolution fashion images with rich structured textual(FIRST) descriptions. In the FIRST, there is a wide range of attire categories and each image-paired textual description is organized at multiple hierarchical levels. Experiments on prevalent generative models trained over FISRT show the necessity of FIRST. We invite the community to further develop more intelligent fashion synthesis and design systems that make fashion design more creative and imaginative based on our dataset. The dataset will be released soon.

Yichen Shi, Ze Zhang, Hongyang Wang et al.

Analog/Mixed-Signal (AMS) circuits play a critical role in the integrated circuit (IC) industry. However, automating Analog/Mixed-Signal (AMS) circuit design has remained a longstanding challenge due to its difficulty and complexity. Although recent advances in Multi-modal Large Language Models (MLLMs) offer promising potential for supporting AMS circuit analysis and design, current research typically evaluates MLLMs on isolated tasks within the domain, lacking a comprehensive benchmark that systematically assesses model capabilities across diverse AMS-related challenges. To address this gap, we introduce AMSbench, a benchmark suite designed to evaluate MLLM performance across critical tasks including circuit schematic perception, circuit analysis, and circuit design. AMSbench comprises approximately 8000 test questions spanning multiple difficulty levels and assesses eight prominent models, encompassing both open-source and proprietary solutions such as Qwen 2.5-VL and Gemini 2.5 Pro. Our evaluation highlights significant limitations in current MLLMs, particularly in complex multi-modal reasoning and sophisticated circuit design tasks. These results underscore the necessity of advancing MLLMs' understanding and effective application of circuit-specific knowledge, thereby narrowing the existing performance gap relative to human expertise and moving toward fully automated AMS circuit design workflows. Our data is released at this URL.

Yiwei Qin, Yixiu Liu, Tiantian Mi et al.

The foundational pretraining phase determines a model's capability ceiling, as post-training struggles to overcome capability foundations established during pretraining, yet it remains critically under-explored. This stems from a structural paradox: organizations with computational resources operate under commercial pressures that inhibit transparent disclosure, while academic institutions possess research freedom but lack pretraining-scale computational resources. daVinci-LLM occupies this unexplored intersection, combining industrial-scale resources with full research freedom to advance the science of pretraining. We adopt a fully-open paradigm that treats openness as scientific methodology, releasing complete data processing pipelines, full training processes, and systematic exploration results. Recognizing that the field lacks systematic methodology for data processing, we employ the Data Darwinism framework, a principled L0-L9 taxonomy from filtering to synthesis. We train a 3B-parameter model from random initialization across 8T tokens using a two-stage adaptive curriculum that progressively shifts from foundational capabilities to reasoning-intensive enhancement. Through 200+ controlled ablations, we establish that: processing depth systematically enhances capabilities, establishing it as a critical dimension alongside volume scaling; different domains exhibit distinct saturation dynamics, necessitating adaptive strategies from proportion adjustments to format shifts; compositional balance enables targeted intensification while preventing performance collapse; how evaluation protocol choices shape our understanding of pretraining progress. By releasing the complete exploration process, we enable the community to build upon our findings and systematic methodologies to form accumulative scientific knowledge in pretraining.

Yunze Wu, Dayuan Fu, Weiye Si et al.

AI agents could accelerate scientific discovery by automating hypothesis formation, experiment design, coding, execution, and analysis, yet existing benchmarks probe narrow skills in simplified settings. To address this gap, we introduce InnovatorBench, a benchmark-platform pair for realistic, end-to-end assessment of agents performing Large Language Model (LLM) research. It comprises 20 tasks spanning Data Construction, Filtering, Augmentation, Loss Design, Reward Design, and Scaffold Construction, which require runnable artifacts and assessment of correctness, performance, output quality, and uncertainty. To support agent operation, we develop ResearchGym, a research environment offering rich action spaces, distributed and long-horizon execution, asynchronous monitoring, and snapshot saving. We also implement a lightweight ReAct agent that couples explicit reasoning with executable planning using frontier models such as Claude-4, GPT-5, GLM-4.5, and Kimi-K2. Our experiments demonstrate that while frontier models show promise in code-driven research tasks, they struggle with fragile algorithm-related tasks and long-horizon decision making, such as impatience, poor resource management, and overreliance on template-based reasoning. Furthermore, agents require over 11 hours to achieve their best performance on InnovatorBench, underscoring the benchmark's difficulty and showing the potential of InnovatorBench to be the next generation of code-based research benchmark.

Yixin Ye, Zhen Huang, Yang Xiao et al.

We challenge the prevailing assumption that complex reasoning in large language models (LLMs) necessitates massive training data. We demonstrate that sophisticated mathematical reasoning can emerge with only a few examples. Specifically, through simple supervised fine-tuning, our model, LIMO, achieves 63.3\% accuracy on AIME24 and 95.6\% on MATH500, surpassing previous fine-tuned models (6.5\% on AIME24, 59.2\% on MATH500) while using only 1\% of the training data required by prior approaches. Furthermore, LIMO exhibits strong out-of-distribution generalization, achieving a 45.8\% absolute improvement across diverse benchmarks, outperforming models trained on 100x more data. Synthesizing these findings, we propose the Less-Is-More Reasoning Hypothesis (LIMO Hypothesis): In foundation models where domain knowledge has been comprehensively encoded during pre-training, sophisticated reasoning can emerge through minimal but strategically designed demonstrations of cognitive processes. This hypothesis suggests that the threshold for eliciting complex reasoning is not dictated by task complexity but rather by two key factors: (1) the completeness of the model's pre-trained knowledge base and (2) the effectiveness of post-training examples in serving as "cognitive templates" that guide reasoning.

Hongbo Ye, Fenghe Tang, Peiang Zhao et al.

Achieving equity in healthcare accessibility requires lightweight yet high-performance solutions for medical image segmentation, particularly in resource-limited settings. Existing methods like U-Net and its variants often suffer from limited global Effective Receptive Fields (ERFs), hindering their ability to capture long-range dependencies. To address this, we propose U-RWKV, a novel framework leveraging the Recurrent Weighted Key-Value(RWKV) architecture, which achieves efficient long-range modeling at O(N) computational cost. The framework introduces two key innovations: the Direction-Adaptive RWKV Module(DARM) and the Stage-Adaptive Squeeze-and-Excitation Module(SASE). DARM employs Dual-RWKV and QuadScan mechanisms to aggregate contextual cues across images, mitigating directional bias while preserving global context and maintaining high computational efficiency. SASE dynamically adapts its architecture to different feature extraction stages, balancing high-resolution detail preservation and semantic relationship capture. Experiments demonstrate that U-RWKV achieves state-of-the-art segmentation performance with high computational efficiency, offering a practical solution for democratizing advanced medical imaging technologies in resource-constrained environments. The code is available at https://github.com/hbyecoding/U-RWKV.

Zhen Huang, Tao Tang, Ronghao Xu et al.

3D landmark detection is a critical task in medical image analysis, and accurately detecting anatomical landmarks is essential for subsequent medical imaging tasks. However, mainstream deep learning methods in this field struggle to simultaneously capture fine-grained local features and model global spatial relationships, while maintaining a balance between accuracy and computational efficiency. Local feature extraction requires capturing fine-grained anatomical details, while global modeling requires understanding the spatial relationships within complex anatomical structures. The high-dimensional nature of 3D volume further exacerbates these challenges, as landmarks are sparsely distributed, leading to significant computational costs. Therefore, achieving efficient and precise 3D landmark detection remains a pressing challenge in medical image analysis. In this work, We propose a \textbf{H}ybrid \textbf{3}D \textbf{DE}tection \textbf{Net}(H3DE-Net), a novel framework that combines CNNs for local feature extraction with a lightweight attention mechanism designed to efficiently capture global dependencies in 3D volumetric data. This mechanism employs a hierarchical routing strategy to reduce computational cost while maintaining global context modeling. To our knowledge, H3DE-Net is the first 3D landmark detection model that integrates such a lightweight attention mechanism with CNNs. Additionally, integrating multi-scale feature fusion further enhances detection accuracy and robustness. Experimental results on a public CT dataset demonstrate that H3DE-Net achieves state-of-the-art(SOTA) performance, significantly improving accuracy and robustness, particularly in scenarios with missing landmarks or complex anatomical variations. We aready open-source our project, including code, data and model weights.

Tiantian Mi, Dongming Shan, Zhen Huang et al.

Data Darwinism (Part I) established a ten-level hierarchy for data processing, showing that stronger processing can unlock greater data value. However, that work relied on manually designed strategies for a single category. Modern pretraining corpora comprise hundreds of heterogeneous categories spanning domains and content types, each demanding specialized treatment. At this scale, manual strategy design becomes prohibitive. This raises a key question: can strategies evolve in an automated way? We introduce DataEvolve, a framework that enables strategies to evolve through iterative optimization rather than manual design. For each data category, DataEvolve operates in a closed evolutionary loop: it identifies quality issues, generates candidate strategies, executes them on sampled data, evaluates results, and refines approaches across generations. The process accumulates knowledge through an experience pool of discovered issues and a strategy pool tracking performance across iterations. Applied to 8 categories spanning 672B tokens from Nemotron-CC, DataEvolve produces Darwin-CC, a 504B-token dataset with strategies evolved through 30 iterations per category. Training 3B models on 500B tokens, Darwin-CC outperforms raw data (+3.96 points) and achieves a 44.13 average score across 18 benchmarks, surpassing DCLM, Ultra-FineWeb, and FineWeb-Edu, with strong gains on knowledge-intensive tasks such as MMLU. Analysis shows evolved strategies converge on cleaning-focused approaches: targeted noise removal and format normalization with domain-aware preservation, echoing the L4 (Generative Refinement) principles from Part I. Ablation studies confirm iterative evolution is essential: optimized strategies outperform suboptimal ones by 2.93 points, establishing evolutionary strategy design as feasible and necessary for pretraining-scale data curation.

Dayuan Fu, Yunze Wu, Xiaojie Cai et al.

Large Language Model (LLM) agents have recently shown strong potential in domains such as automated coding, deep research, and graphical user interface manipulation. However, training them to succeed on long-horizon, domain-specialized tasks remains challenging. Current methods primarily fall into two categories. The first relies on dense human annotations through behavior cloning, which is prohibitively expensive for long-horizon tasks that can take days or months. The second depends on outcome-driven sampling, which often collapses due to the rarity of valid positive trajectories on domain-specialized tasks. We introduce Apollo, a sampling framework that integrates asynchronous human guidance with action-level data filtering. Instead of requiring annotators to shadow every step, Apollo allows them to intervene only when the agent drifts from a promising trajectory, by providing prior knowledge, strategic advice, etc. This lightweight design makes it possible to sustain interactions for over 30 hours and produces valuable trajectories at a lower cost. Apollo then applies supervision control to filter out sub-optimal actions and prevent error propagation. Together, these components enable reliable and effective data collection in long-horizon environments. To demonstrate the effectiveness of Apollo, we evaluate it using InnovatorBench. Our experiments show that when applied to train the GLM-4.5 model on InnovatorBench, Apollo achieves more than a 50% improvement over the untrained baseline and a 28% improvement over a variant trained without human interaction. These results highlight the critical role of human-in-the-loop sampling and the robustness of Apollo's design in handling long-horizon, domain-specialized tasks.

Zhen Huang, Jiaxin Deng, Jiayu Xu et al.

In bus arrival time prediction, the process of organizing road infrastructure network data into homogeneous entities is known as segmentation. Segmenting a road network is widely recognized as the first and most critical step in developing an arrival time prediction system, particularly for auto-regressive-based approaches. Traditional methods typically employ a uniform segmentation strategy, which fails to account for varying physical constraints along roads, such as road conditions, intersections, and points of interest, thereby limiting prediction efficiency. In this paper, we propose a Reinforcement Learning (RL)-based approach to efficiently and adaptively learn non-uniform road segments for arrival time prediction. Our method decouples the prediction process into two stages: 1) Non-uniform road segments are extracted based on their impact scores using the proposed RL framework; and 2) A linear prediction model is applied to the selected segments to make predictions. This method ensures optimal segment selection while maintaining computational efficiency, offering a significant improvement over traditional uniform approaches. Furthermore, our experimental results suggest that the linear approach can even achieve better performance than more complex methods. Extensive experiments demonstrate the superiority of the proposed method, which not only enhances efficiency but also improves learning performance on large-scale benchmarks. The dataset and the code are publicly accessible at: https://github.com/pangjunbiao/Less-is-More.

Rongsheng Wang, Fenghe Tang, Qingsong Yao et al.

Medical vision-language pre-training shows great potential in learning representative features from massive paired radiographs and reports. However, in computed tomography (CT) scans, the distribution of lesions which contain intricate structures is characterized by spatial sparsity. Besides, the complex and implicit relationships between different pathological descriptions in each sentence of the report and their corresponding sub-regions in radiographs pose additional challenges. In this paper, we propose a Similarity-Driven Cross-Granularity Pre-training (SimCroP) framework on chest CTs, which combines similarity-driven alignment and cross-granularity fusion to improve radiograph interpretation. We first leverage multi-modal masked modeling to optimize the encoder for understanding precise low-level semantics from radiographs. Then, similarity-driven alignment is designed to pre-train the encoder to adaptively select and align the correct patches corresponding to each sentence in reports. The cross-granularity fusion module integrates multimodal information across instance level and word-patch level, which helps the model better capture key pathology structures in sparse radiographs, resulting in improved performance for multi-scale downstream tasks. SimCroP is pre-trained on a large-scale paired CT-reports dataset and validated on image classification and segmentation tasks across five public datasets. Experimental results demonstrate that SimCroP outperforms both cutting-edge medical self-supervised learning methods and medical vision-language pre-training methods. Codes and models are available at https://github.com/ToniChopp/SimCroP.

Zhen Huang, Zengzhi Wang, Shijie Xia et al.

In this report, we pose the following question: Who is the most intelligent AI model to date, as measured by the OlympicArena (an Olympic-level, multi-discipline, multi-modal benchmark for superintelligent AI)? We specifically focus on the most recently released models: Claude-3.5-Sonnet, Gemini-1.5-Pro, and GPT-4o. For the first time, we propose using an Olympic medal Table approach to rank AI models based on their comprehensive performance across various disciplines. Empirical results reveal: (1) Claude-3.5-Sonnet shows highly competitive overall performance over GPT-4o, even surpassing GPT-4o on a few subjects (i.e., Physics, Chemistry, and Biology). (2) Gemini-1.5-Pro and GPT-4V are ranked consecutively just behind GPT-4o and Claude-3.5-Sonnet, but with a clear performance gap between them. (3) The performance of AI models from the open-source community significantly lags behind these proprietary models. (4) The performance of these models on this benchmark has been less than satisfactory, indicating that we still have a long way to go before achieving superintelligence. We remain committed to continuously tracking and evaluating the performance of the latest powerful models on this benchmark (available at https://github.com/GAIR-NLP/OlympicArena).

Zhen Huang, Zengzhi Wang, Shijie Xia et al.

The evolution of Artificial Intelligence (AI) has been significantly accelerated by advancements in Large Language Models (LLMs) and Large Multimodal Models (LMMs), gradually showcasing potential cognitive reasoning abilities in problem-solving and scientific discovery (i.e., AI4Science) once exclusive to human intellect. To comprehensively evaluate current models' performance in cognitive reasoning abilities, we introduce OlympicArena, which includes 11,163 bilingual problems across both text-only and interleaved text-image modalities. These challenges encompass a wide range of disciplines spanning seven fields and 62 international Olympic competitions, rigorously examined for data leakage. We argue that the challenges in Olympic competition problems are ideal for evaluating AI's cognitive reasoning due to their complexity and interdisciplinary nature, which are essential for tackling complex scientific challenges and facilitating discoveries. Beyond evaluating performance across various disciplines using answer-only criteria, we conduct detailed experiments and analyses from multiple perspectives. We delve into the models' cognitive reasoning abilities, their performance across different modalities, and their outcomes in process-level evaluations, which are vital for tasks requiring complex reasoning with lengthy solutions. Our extensive evaluations reveal that even advanced models like GPT-4o only achieve a 39.97% overall accuracy, illustrating current AI limitations in complex reasoning and multimodal integration. Through the OlympicArena, we aim to advance AI towards superintelligence, equipping it to address more complex challenges in science and beyond. We also provide a comprehensive set of resources to support AI research, including a benchmark dataset, an open-source annotation platform, a detailed evaluation tool, and a leaderboard with automatic submission features.

DeepSeek-AI, Aixin Liu, Bei Feng et al. · stanford, tsinghua

We present DeepSeek-V3, a strong Mixture-of-Experts (MoE) language model with 671B total parameters with 37B activated for each token. To achieve efficient inference and cost-effective training, DeepSeek-V3 adopts Multi-head Latent Attention (MLA) and DeepSeekMoE architectures, which were thoroughly validated in DeepSeek-V2. Furthermore, DeepSeek-V3 pioneers an auxiliary-loss-free strategy for load balancing and sets a multi-token prediction training objective for stronger performance. We pre-train DeepSeek-V3 on 14.8 trillion diverse and high-quality tokens, followed by Supervised Fine-Tuning and Reinforcement Learning stages to fully harness its capabilities. Comprehensive evaluations reveal that DeepSeek-V3 outperforms other open-source models and achieves performance comparable to leading closed-source models. Despite its excellent performance, DeepSeek-V3 requires only 2.788M H800 GPU hours for its full training. In addition, its training process is remarkably stable. Throughout the entire training process, we did not experience any irrecoverable loss spikes or perform any rollbacks. The model checkpoints are available at https://github.com/deepseek-ai/DeepSeek-V3.

Xiaoqian Zhou, Zhen Huang, Heqin Zhu et al.

Anatomical landmark detection (ALD) from a medical image is crucial for a wide array of clinical applications. While existing methods achieve quite some success in ALD, they often struggle to balance global context with computational efficiency, particularly with high-resolution images, thereby leading to the rise of a natural question: where is the performance limit of ALD? In this paper, we aim to forge performant ALD by proposing a {\bf HY}brid {\bf ATT}ention {\bf Net}work (HYATT-Net) with the following designs: (i) A novel hybrid architecture that integrates CNNs and Transformers. Its core is the BiFormer module, utilizing Bi-Level Routing Attention for efficient attention to relevant image regions. This, combined with Attention Residual Module(ARM), enables precise local feature refinement guided by the global context. (ii) A Feature Fusion Correction Module that aggregates multi-scale features and thus mitigates a resolution loss. Deep supervision with a mean-square error loss on multi-resolution heatmaps optimizes the model. Experiments on five diverse datasets demonstrate state-of-the-art performance, surpassing existing methods in accuracy, robustness, and efficiency. The HYATT-Net provides a promising solution for accurate and efficient ALD in complex medical images. Our codes and data are already released at: \url{https://github.com/ECNUACRush/HYATT-Net}.

Zhen Huang, Xu Shen, Jun Xing et al.

Knowledge Distillation (KD) is a popular technique to transfer knowledge from a teacher model or ensemble to a student model. Its success is generally attributed to the privileged information on similarities/consistency between the class distributions or intermediate feature representations of the teacher model and the student model. However, directly pushing the student model to mimic the probabilities/features of the teacher model to a large extent limits the student model in learning undiscovered knowledge/features. In this paper, we propose a novel inheritance and exploration knowledge distillation framework (IE-KD), in which a student model is split into two parts - inheritance and exploration. The inheritance part is learned with a similarity loss to transfer the existing learned knowledge from the teacher model to the student model, while the exploration part is encouraged to learn representations different from the inherited ones with a dis-similarity loss. Our IE-KD framework is generic and can be easily combined with existing distillation or mutual learning methods for training deep neural networks. Extensive experiments demonstrate that these two parts can jointly push the student model to learn more diversified and effective representations, and our IE-KD can be a general technique to improve the student network to achieve SOTA performance. Furthermore, by applying our IE-KD to the training of two networks, the performance of both can be improved w.r.t. deep mutual learning. The code and models of IE-KD will be make publicly available at https://github.com/yellowtownhz/IE-KD.

Xin Jin, Tianyu He, Kecheng Zheng et al.

Cloth-Changing person re-identification (CC-ReID) aims at matching the same person across different locations over a long-duration, e.g., over days, and therefore inevitably meets challenge of changing clothing. In this paper, we focus on handling well the CC-ReID problem under a more challenging setting, i.e., just from a single image, which enables high-efficiency and latency-free pedestrian identify for real-time surveillance applications. Specifically, we introduce Gait recognition as an auxiliary task to drive the Image ReID model to learn cloth-agnostic representations by leveraging personal unique and cloth-independent gait information, we name this framework as GI-ReID. GI-ReID adopts a two-stream architecture that consists of a image ReID-Stream and an auxiliary gait recognition stream (Gait-Stream). The Gait-Stream, that is discarded in the inference for high computational efficiency, acts as a regulator to encourage the ReID-Stream to capture cloth-invariant biometric motion features during the training. To get temporal continuous motion cues from a single image, we design a Gait Sequence Prediction (GSP) module for Gait-Stream to enrich gait information. Finally, a high-level semantics consistency over two streams is enforced for effective knowledge regularization. Experiments on multiple image-based Cloth-Changing ReID benchmarks, e.g., LTCC, PRCC, Real28, and VC-Clothes, demonstrate that GI-ReID performs favorably against the state-of-the-arts. Codes are available at https://github.com/jinx-USTC/GI-ReID.

Zhen Huang, Xu Shen, Xinmei Tian et al.

Skeleton-based human action recognition has attracted much attention with the prevalence of accessible depth sensors. Recently, graph convolutional networks (GCNs) have been widely used for this task due to their powerful capability to model graph data. The topology of the adjacency graph is a key factor for modeling the correlations of the input skeletons. Thus, previous methods mainly focus on the design/learning of the graph topology. But once the topology is learned, only a single-scale feature and one transformation exist in each layer of the networks. Many insights, such as multi-scale information and multiple sets of transformations, that have been proven to be very effective in convolutional neural networks (CNNs), have not been investigated in GCNs. The reason is that, due to the gap between graph-structured skeleton data and conventional image/video data, it is very challenging to embed these insights into GCNs. To overcome this gap, we reinvent the split-transform-merge strategy in GCNs for skeleton sequence processing. Specifically, we design a simple and highly modularized graph convolutional network architecture for skeleton-based action recognition. Our network is constructed by repeating a building block that aggregates multi-granularity information from both the spatial and temporal paths. Extensive experiments demonstrate that our network outperforms state-of-the-art methods by a significant margin with only 1/5 of the parameters and 1/10 of the FLOPs. Code is available at https://github.com/yellowtownhz/STIGCN.

Minghao Hu, Yuxing Peng, Zhen Huang et al.

Rapid progress has been made in the field of reading comprehension and question answering, where several systems have achieved human parity in some simplified settings. However, the performance of these models degrades significantly when they are applied to more realistic scenarios, such as answers involve various types, multiple text strings are correct answers, or discrete reasoning abilities are required. In this paper, we introduce the Multi-Type Multi-Span Network (MTMSN), a neural reading comprehension model that combines a multi-type answer predictor designed to support various answer types (e.g., span, count, negation, and arithmetic expression) with a multi-span extraction method for dynamically producing one or multiple text spans. In addition, an arithmetic expression reranking mechanism is proposed to rank expression candidates for further confirming the prediction. Experiments show that our model achieves 79.9 F1 on the DROP hidden test set, creating new state-of-the-art results. Source code\footnote{\url{https://github.com/huminghao16/MTMSN}} is released to facilitate future work.