Ziang Li

Qijia He, Xunmei Liu, Hammaad Memon et al. · allen-ai

Scalable Vector Graphics (SVG) are an essential format for technical illustration and digital design, offering precise resolution independence and flexible semantic editability. In practice, however, original vector source files are frequently lost or inaccessible, leaving only "flat" rasterized versions (e.g., PNG or JPEG) that are difficult to modify or scale. Manually reconstructing these figures is a prohibitively labor-intensive process, requiring specialized expertise to recover the original geometric intent. To bridge this gap, we propose VFIG, a family of Vision-Language Models trained for complex and high-fidelity figure-to-SVG conversion. While this task is inherently data-driven, existing datasets are typically small-scale and lack the complexity of professional diagrams. We address this by introducing VFIG-DATA, a large-scale dataset of 66K high-quality figure-SVG pairs, curated from a diverse mix of real-world paper figures and procedurally generated diagrams. Recognizing that SVGs are composed of recurring primitives and hierarchical local structures, we introduce a coarse-to-fine training curriculum that begins with supervised fine-tuning (SFT) to learn atomic primitives and transitions to reinforcement learning (RL) refinement to optimize global diagram fidelity, layout consistency, and topological edge cases. Finally, we introduce VFIG-BENCH, a comprehensive evaluation suite with novel metrics designed to measure the structural integrity of complex figures. VFIG achieves state-of-the-art performance among open-source models and performs on par with GPT-5.2, achieving a VLM-Judge score of 0.829 on VFIG-BENCH.

Kevin Yu, Jihye Roh, Ziang Li et al.

Computer-aided synthesis planning (CASP) algorithms have demonstrated expert-level abilities in planning retrosynthetic routes to molecules of low to moderate complexity. However, current search methods assume the sufficiency of reaching arbitrary building blocks, failing to address the common real-world constraint where using specific molecules is desired. To this end, we present a formulation of synthesis planning with starting material constraints. Under this formulation, we propose Double-Ended Synthesis Planning (DESP), a novel CASP algorithm under a bidirectional graph search scheme that interleaves expansions from the target and from the goal starting materials to ensure constraint satisfiability. The search algorithm is guided by a goal-conditioned cost network learned offline from a partially observed hypergraph of valid chemical reactions. We demonstrate the utility of DESP in improving solve rates and reducing the number of search expansions by biasing synthesis planning towards expert goals on multiple new benchmarks. DESP can make use of existing one-step retrosynthesis models, and we anticipate its performance to scale as these one-step model capabilities improve.

Ziang Li, Ming Ding, Weikai Li et al. · tsinghua

We argue that the present setting of semisupervised learning on graphs may result in unfair comparisons, due to its potential risk of over-tuning hyper-parameters for models. In this paper, we highlight the significant influence of tuning hyper-parameters, which leverages the label information in the validation set to improve the performance. To explore the limit of over-tuning hyperparameters, we propose ValidUtil, an approach to fully utilize the label information in the validation set through an extra group of hyper-parameters. With ValidUtil, even GCN can easily get high accuracy of 85.8% on Cora. To avoid over-tuning, we merge the training set and the validation set and construct an i.i.d. graph benchmark (IGB) consisting of 4 datasets. Each dataset contains 100 i.i.d. graphs sampled from a large graph to reduce the evaluation variance. Our experiments suggest that IGB is a more stable benchmark than previous datasets for semisupervised learning on graphs.

Hang Yin, Pinren Lu, Ziang Li et al.

The need for high-quality data has been a key issue hindering the research of dialogue tasks. Recent studies try to build datasets through manual, web crawling, and large pre-trained models. However, man-made data is expensive and data collected from the internet often includes generic responses, meaningless statements, and toxic dialogues. Automatic data generation through large models is a cost-effective method, but for open-domain multimodal dialogue tasks, there are still three drawbacks: 1) There is currently no open-source large model that can accept multimodal input; 2) The content generated by the model lacks interpretability; 3) The generated data is usually difficult to quality control and require extensive resource to collect. To alleviate the significant human and resource expenditure in data collection, we propose a Multimodal Data Construction Framework (MDCF). MDCF designs proper prompts to spur the large-scale pre-trained language model to generate well-formed and satisfactory content. Additionally, MDCF also automatically provides explanation for a given image and its corresponding dialogue, which can provide a certain degree of interpretability and facilitate manual follow-up quality inspection. Based on this, we release an Explanatory Multimodal Open-Domain dialogue dataset (EXMODD). Experiments indicate a positive correlation between the model's ability to generate accurate understandings and high-quality responses. Our code and data can be found at https://github.com/poplpr/EXMODD.

Shaoke Fang, Ziang Li, Wenfei Wu et al.

Prefix caching is a key optimization in Large Language Model (LLM) serving, reusing attention Key-Value (KV) states across requests with shared prompt prefixes to reduce expensive prefill computation. However, its benefit depends critically on the eviction policy as GPU memory is scarce, and existing policies such as LRU largely treat cached blocks uniformly. This view ignores a fundamental property of LLM prompts: not all tokens are equally worth caching. We show that different token types within a prompt, including system prompts, user queries, tool outputs, model responses, and chain-of-thought reasoning, exhibit up to 756x variation in reuse rates, yet no existing eviction policy exploits this signal. In this paper, we present SAECache (Semantic-Adaptive Eviction for prefix caches), a semantic-adaptive prefix cache eviction policy that addresses this gap through three innovations: (1) a multi-queue architecture that routes KV blocks to task-specific queues with tailored priority metrics, capturing both session reuse in multi-turn requests and structural reuse in templated single-turn requests; (2) a semantic-aware token weighting mechanism that learns the reuse value of different token types online through eviction feedback; and (3) a fully adaptive online learning schema for all parameter updates, including log-normal timing parameters, position decay power, queue weights, and meta-parameters, which eliminates manual tuning and enables automatic adaptation to deployment-specific workload characteristics. Through extensive evaluation across heterogeneous workloads, we demonstrate that SAECache achieves 1.4x-2.7x TTFT improvement over production-style baselines, while fixed-parameter alternatives can degrade by up to 2.7x under workload mismatch -- a failure mode our adaptive approach avoids entirely.

Yijia Xiao, Jiezhong Qiu, Ziang Li et al.

Protein is linked to almost every life process. Therefore, analyzing the biological structure and property of protein sequences is critical to the exploration of life, as well as disease detection and drug discovery. Traditional protein analysis methods tend to be labor-intensive and time-consuming. The emergence of deep learning models makes modeling data patterns in large quantities of data possible. Interdisciplinary researchers have begun to leverage deep learning methods to model large biological datasets, e.g. using long short-term memory and convolutional neural network for protein sequence classification. After millions of years of evolution, evolutionary information is encoded in protein sequences. Inspired by the similarity between natural language and protein sequences, we use large-scale language models to model evolutionary-scale protein sequences, encoding protein biology information in representation. Significant improvements are observed in both token-level and sequence-level tasks, demonstrating that our large-scale model can accurately capture evolution information from pretraining on evolutionary-scale individual sequences. Our code and model are available at https://github.com/THUDM/ProteinLM.

Linjie Li, Mahtab Bigverdi, Jiawei Gu et al.

Spatial cognition is essential for human intelligence, enabling problem-solving through visual simulations rather than solely relying on verbal reasoning. However, existing AI benchmarks primarily assess verbal reasoning, neglecting the complexities of non-verbal, multi-step visual simulation. We introduce STARE(Spatial Transformations and Reasoning Evaluation), a benchmark designed to rigorously evaluate multimodal large language models on tasks better solved through multi-step visual simulation. STARE features 4K tasks spanning foundational geometric transformations (2D and 3D), integrated spatial reasoning (cube net folding and tangram puzzles), and real-world spatial reasoning (perspective and temporal reasoning), reflecting practical cognitive challenges like object assembly, mechanical diagram interpretation, and everyday spatial navigation. Our evaluations show that models excel at reasoning over simpler 2D transformations, but perform close to random chance on more complex tasks like 3D cube net folding and tangram puzzles that require multi-step visual simulations. Humans achieve near-perfect accuracy but take considerable time (up to 28.9s) on complex tasks, significantly speeding up (down by 7.5 seconds on average) with intermediate visual simulations. In contrast, models exhibit inconsistent performance gains from visual simulations, improving on most tasks but declining in specific cases like tangram puzzles (GPT-4o, o1) and cube net folding (Claude-3.5, Gemini-2.0 Flash), indicating that models may not know how to effectively leverage intermediate visual information.

Peng Xu, Shengwu Xiong, Jiajun Zhang et al.

This paper reviews the MARS2 2025 Challenge on Multimodal Reasoning. We aim to bring together different approaches in multimodal machine learning and LLMs via a large benchmark. We hope it better allows researchers to follow the state-of-the-art in this very dynamic area. Meanwhile, a growing number of testbeds have boosted the evolution of general-purpose large language models. Thus, this year's MARS2 focuses on real-world and specialized scenarios to broaden the multimodal reasoning applications of MLLMs. Our organizing team released two tailored datasets Lens and AdsQA as test sets, which support general reasoning in 12 daily scenarios and domain-specific reasoning in advertisement videos, respectively. We evaluated 40+ baselines that include both generalist MLLMs and task-specific models, and opened up three competition tracks, i.e., Visual Grounding in Real-world Scenarios (VG-RS), Visual Question Answering with Spatial Awareness (VQA-SA), and Visual Reasoning in Creative Advertisement Videos (VR-Ads). Finally, 76 teams from the renowned academic and industrial institutions have registered and 40+ valid submissions (out of 1200+) have been included in our ranking lists. Our datasets, code sets (40+ baselines and 15+ participants' methods), and rankings are publicly available on the MARS2 workshop website and our GitHub organization page https://github.com/mars2workshop/, where our updates and announcements of upcoming events will be continuously provided.

Ziang Li, Manasi Ganti, Zixian Ma et al.

Large language models (LLMs) often generate natural language rationales -- free-form explanations that help improve performance on complex reasoning tasks and enhance interpretability for human users. However, evaluating these rationales remains challenging. While recent work has relied on binary preference judgments from humans or LLM judges, such evaluations are often opaque and coarse-grained, offering limited insight into what makes one rationale better than another. In this work, we rethink preference evaluation for LLM-generated rationales by asking: (1) What attributes define good rationales? (2) Can human preferences be explained by these attributes? (3) Can attribute-based evaluation overcome the limitations of binary comparisons? We identify a set of key rationale attributes from prior literature and assess them using automatic metrics, LLM judgments, and human annotations. We then analyze two standard human preference datasets MT Bench and Chatbot Arena using SHAP to identify which attributes best explain human preference outcomes. Finally, we re-evaluate model-generated rationales using attribute-specific ELO scores, revealing more nuanced model comparisons and insights. Our findings suggest that fine-grained attribute evaluations can better characterize rationale quality and guide future research toward more interpretable and reliable evaluation practices.

Jack Erhardt, Ziang Li, Reid Pinkham et al.

Machine learning algorithms have enabled high quality stereo depth estimation to run on Augmented and Virtual Reality (AR/VR) devices. However, high energy consumption across the full image processing stack prevents stereo depth algorithms from running effectively on battery-limited devices. This paper introduces SteROI-D, a full stereo depth system paired with a mapping methodology. SteROI-D exploits Region-of-Interest (ROI) and temporal sparsity at the system level to save energy. SteROI-D's flexible and heterogeneous compute fabric supports diverse ROIs. Importantly, we introduce a systematic mapping methodology to effectively handle dynamic ROIs, thereby maximizing energy savings. Using these techniques, our 28nm prototype SteROI-D design achieves up to 4.35x reduction in total system energy compared to a baseline ASIC.

Ziang Li, Yiwen Guo, Haodi Liu et al.

Backpropagation (BP) is widely used for calculating gradients in deep neural networks (DNNs). Applied often along with stochastic gradient descent (SGD) or its variants, BP is considered as a de-facto choice in a variety of machine learning tasks including DNN training and adversarial attack/defense. Recently, a linear variant of BP named LinBP was introduced for generating more transferable adversarial examples for performing black-box attacks, by Guo et al. Although it has been shown empirically effective in black-box attacks, theoretical studies and convergence analyses of such a method is lacking. This paper serves as a complement and somewhat an extension to Guo et al.'s paper, by providing theoretical analyses on LinBP in neural-network-involved learning tasks, including adversarial attack and model training. We demonstrate that, somewhat surprisingly, LinBP can lead to faster convergence in these tasks in the same hyper-parameter settings, compared to BP. We confirm our theoretical results with extensive experiments.

Ziang Li, Kailun Wu, Yiwen Guo et al.

Drawing on theoretical insights, we advocate an error-based thresholding (EBT) mechanism for learned ISTA (LISTA), which utilizes a function of the layer-wise reconstruction error to suggest a specific threshold for each observation in the shrinkage function of each layer. We show that the proposed EBT mechanism well disentangles the learnable parameters in the shrinkage functions from the reconstruction errors, endowing the obtained models with improved adaptivity to possible data variations. With rigorous analyses, we further show that the proposed EBT also leads to a faster convergence on the basis of LISTA or its variants, in addition to its higher adaptivity. Extensive experimental results confirm our theoretical analyses and verify the effectiveness of our methods.

Dewei Liu, Chuan He, Xin Peng et al.

Availability issues of industrial microservice systems (e.g., drop of successfully placed orders and processed transactions) directly affect the running of the business. These issues are usually caused by various types of service anomalies which propagate along service dependencies. Accurate and high-efficient root cause localization is thus a critical challenge for large-scale industrial microservice systems. Existing approaches use service dependency graph based analysis techniques to automatically locate root causes. However, these approaches are limited due to their inaccurate detection of service anomalies and inefficient traversing of service dependency graph. In this paper, we propose a high-efficient root cause localization approach for availability issues of microservice systems, called MicroHECL. Based on a dynamically constructed service call graph, MicroHECL analyzes possible anomaly propagation chains, and ranks candidate root causes based on correlation analysis. We combine machine learning and statistical methods and design customized models for the detection of different types of service anomalies (i.e., performance, reliability, traffic). To improve the efficiency, we adopt a pruning strategy to eliminate irrelevant service calls in anomaly propagation chain analysis. Experimental studies show that MicroHECL significantly outperforms two state-of-the-art baseline approaches in terms of both accuracy and efficiency. MicroHECL has been used in Alibaba and achieves a top-3 hit ratio of 68% with root cause localization time reduced from 30 minutes to 5 minutes.