Tianyu Liu

Xiangru Tang, Yuliang Liu, Zefan Cai et al. · pku

Despite Large Language Models (LLMs) like GPT-4 achieving impressive results in function-level code generation, they struggle with repository-scale code understanding (e.g., coming up with the right arguments for calling routines), requiring a deeper comprehension of complex file interactions. Also, recently, people have developed LLM agents that attempt to interact with repository code (e.g., compiling and evaluating its execution), prompting the need to evaluate their performance. These gaps have motivated our development of ML-Bench, a benchmark rooted in real-world programming applications that leverage existing code repositories to perform tasks. Addressing the need for LLMs to interpret long code contexts and translate instructions into precise, executable scripts, ML-Bench encompasses annotated 9,641 examples across 18 GitHub repositories, challenging LLMs to accommodate user-specified arguments and documentation intricacies effectively. To evaluate both LLMs and AI agents, two setups are employed: ML-LLM-Bench for assessing LLMs' text-to-code conversion within a predefined deployment environment, and ML-Agent-Bench for testing autonomous agents in an end-to-end task execution within a Linux sandbox environment. Our findings indicate that while GPT-4o leads with a Pass@5 rate surpassing 50%, there remains significant scope for improvement, highlighted by issues such as hallucinated outputs and difficulties with bash script generation. Notably, in the more demanding ML-Agent-Bench, GPT-4o achieves a 76.47% success rate, reflecting the efficacy of iterative action and feedback in complex task resolution. Our code, dataset, and models are available at https://github.com/gersteinlab/ML-bench.

Yifan Song, Peiyi Wang, Weimin Xiong et al. · pku

Continual learning (CL) aims to constantly learn new knowledge over time while avoiding catastrophic forgetting on old tasks. We focus on continual text classification under the class-incremental setting. Recent CL studies have identified the severe performance decrease on analogous classes as a key factor for catastrophic forgetting. In this paper, through an in-depth exploration of the representation learning process in CL, we discover that the compression effect of the information bottleneck leads to confusion on analogous classes. To enable the model learn more sufficient representations, we propose a novel replay-based continual text classification method, InfoCL. Our approach utilizes fast-slow and current-past contrastive learning to perform mutual information maximization and better recover the previously learned representations. In addition, InfoCL incorporates an adversarial memory augmentation strategy to alleviate the overfitting problem of replay. Experimental results demonstrate that InfoCL effectively mitigates forgetting and achieves state-of-the-art performance on three text classification tasks. The code is publicly available at https://github.com/Yifan-Song793/InfoCL.

Liang Chen, Yichi Zhang, Shuhuai Ren et al. · pku

In this study, we explore the potential of Multimodal Large Language Models (MLLMs) in improving embodied decision-making processes for agents. While Large Language Models (LLMs) have been widely used due to their advanced reasoning skills and vast world knowledge, MLLMs like GPT4-Vision offer enhanced visual understanding and reasoning capabilities. We investigate whether state-of-the-art MLLMs can handle embodied decision-making in an end-to-end manner and whether collaborations between LLMs and MLLMs can enhance decision-making. To address these questions, we introduce a new benchmark called PCA-EVAL, which evaluates embodied decision-making from the perspectives of Perception, Cognition, and Action. Additionally, we propose HOLMES, a multi-agent cooperation framework that allows LLMs to leverage MLLMs and APIs to gather multimodal information for informed decision-making. We compare end-to-end embodied decision-making and HOLMES on our benchmark and find that the GPT4-Vision model demonstrates strong end-to-end embodied decision-making abilities, outperforming GPT4-HOLMES in terms of average decision accuracy (+3%). However, this performance is exclusive to the latest GPT4-Vision model, surpassing the open-source state-of-the-art MLLM by 26%. Our results indicate that powerful MLLMs like GPT4-Vision hold promise for decision-making in embodied agents, offering new avenues for MLLM research. Code and data are open at https://github.com/pkunlp-icler/PCA-EVAL/.

Zirui Wu, Tianyu Liu, Liyi Luo et al.

Nowadays, autonomous cars can drive smoothly in ordinary cases, and it is widely recognized that realistic sensor simulation will play a critical role in solving remaining corner cases by simulating them. To this end, we propose an autonomous driving simulator based upon neural radiance fields (NeRFs). Compared with existing works, ours has three notable features: (1) Instance-aware. Our simulator models the foreground instances and background environments separately with independent networks so that the static (e.g., size and appearance) and dynamic (e.g., trajectory) properties of instances can be controlled separately. (2) Modular. Our simulator allows flexible switching between different modern NeRF-related backbones, sampling strategies, input modalities, etc. We expect this modular design to boost academic progress and industrial deployment of NeRF-based autonomous driving simulation. (3) Realistic. Our simulator set new state-of-the-art photo-realism results given the best module selection. Our simulator will be open-sourced while most of our counterparts are not. Project page: https://open-air-sun.github.io/mars/.

Zihan Wang, Peiyi Wang, Tianyu Liu et al.

Hierarchical text classification (HTC) is a challenging subtask of multi-label classification due to its complex label hierarchy. Recently, the pretrained language models (PLM)have been widely adopted in HTC through a fine-tuning paradigm. However, in this paradigm, there exists a huge gap between the classification tasks with sophisticated label hierarchy and the masked language model (MLM) pretraining tasks of PLMs and thus the potentials of PLMs can not be fully tapped. To bridge the gap, in this paper, we propose HPT, a Hierarchy-aware Prompt Tuning method to handle HTC from a multi-label MLM perspective. Specifically, we construct a dynamic virtual template and label words that take the form of soft prompts to fuse the label hierarchy knowledge and introduce a zero-bounded multi-label cross entropy loss to harmonize the objectives of HTC and MLM. Extensive experiments show HPT achieves state-of-the-art performances on 3 popular HTC datasets and is adept at handling the imbalance and low resource situations. Our code is available at https://github.com/wzh9969/HPT.

Runxin Xu, Peiyi Wang, Tianyu Liu et al. · pku

Most previous studies aim at extracting events from a single sentence, while document-level event extraction still remains under-explored. In this paper, we focus on extracting event arguments from an entire document, which mainly faces two critical problems: a) the long-distance dependency between trigger and arguments over sentences; b) the distracting context towards an event in the document. To address these issues, we propose a Two-Stream Abstract meaning Representation enhanced extraction model (TSAR). TSAR encodes the document from different perspectives by a two-stream encoding module, to utilize local and global information and lower the impact of distracting context. Besides, TSAR introduces an AMR-guided interaction module to capture both intra-sentential and inter-sentential features, based on the locally and globally constructed AMR semantic graphs. An auxiliary boundary loss is introduced to enhance the boundary information for text spans explicitly. Extensive experiments illustrate that TSAR outperforms previous state-of-the-art by a large margin, with 2.54 F1 and 5.13 F1 performance gain on the public RAMS and WikiEvents datasets respectively, showing the superiority in the cross-sentence arguments extraction. We release our code in https://github.com/ PKUnlp-icler/TSAR.

Peiyi Wang, Yifan Song, Tianyu Liu et al. · pku

Continual relation extraction (CRE) aims to continually learn new relations from a class-incremental data stream. CRE model usually suffers from catastrophic forgetting problem, i.e., the performance of old relations seriously degrades when the model learns new relations. Most previous work attributes catastrophic forgetting to the corruption of the learned representations as new relations come, with an implicit assumption that the CRE models have adequately learned the old relations. In this paper, through empirical studies we argue that this assumption may not hold, and an important reason for catastrophic forgetting is that the learned representations do not have good robustness against the appearance of analogous relations in the subsequent learning process. To address this issue, we encourage the model to learn more precise and robust representations through a simple yet effective adversarial class augmentation mechanism (ACA), which is easy to implement and model-agnostic. Experimental results show that ACA can consistently improve the performance of state-of-the-art CRE models on two popular benchmarks.

Binyuan Hui, Jian Yang, Zeyu Cui et al.

In this report, we introduce the Qwen2.5-Coder series, a significant upgrade from its predecessor, CodeQwen1.5. This series includes six models: Qwen2.5-Coder-(0.5B/1.5B/3B/7B/14B/32B). As a code-specific model, Qwen2.5-Coder is built upon the Qwen2.5 architecture and continues pretrained on a vast corpus of over 5.5 trillion tokens. Through meticulous data cleaning, scalable synthetic data generation, and balanced data mixing, Qwen2.5-Coder demonstrates impressive code generation capabilities while retaining general and math skills. These models have been evaluated on a wide range of code-related tasks, achieving state-of-the-art (SOTA) performance across more than 10 benchmarks, including code generation, completion, reasoning, and repair, consistently outperforming larger models of the same model size. We believe that the release of the Qwen2.5-Coder series will advance research in code intelligence and, with its permissive licensing, support wider adoption by developers in real-world applications.

Clément Guerner, Tianyu Liu, Anej Svete et al. · allen-ai, eth-zurich

The linear subspace hypothesis (Bolukbasi et al., 2016) states that, in a language model's representation space, all information about a concept such as verbal number is encoded in a linear subspace. Prior work has relied on auxiliary classification tasks to identify and evaluate candidate subspaces that might give support for this hypothesis. We instead give a set of intrinsic criteria which characterize an ideal linear concept subspace and enable us to identify the subspace using only the language model distribution. Our information-theoretic framework accounts for spuriously correlated features in the representation space (Kumar et al., 2022) by reconciling the statistical notion of concept information and the geometric notion of how concepts are encoded in the representation space. As a byproduct of this analysis, we hypothesize a causal process for how a language model might leverage concepts during generation. Empirically, we find that linear concept erasure is successful in erasing most concept information under our framework for verbal number as well as some complex aspect-level sentiment concepts from a restaurant review dataset. Our causal intervention for controlled generation shows that, for at least one concept across two languages models, the concept subspace can be used to manipulate the concept value of the generated word with precision.

Liang Chen, Peiyi Wang, Runxin Xu et al. · pku

As Abstract Meaning Representation (AMR) implicitly involves compound semantic annotations, we hypothesize auxiliary tasks which are semantically or formally related can better enhance AMR parsing. We find that 1) Semantic role labeling (SRL) and dependency parsing (DP), would bring more performance gain than other tasks e.g. MT and summarization in the text-to-AMR transition even with much less data. 2) To make a better fit for AMR, data from auxiliary tasks should be properly "AMRized" to PseudoAMR before training. Knowledge from shallow level parsing tasks can be better transferred to AMR Parsing with structure transform. 3) Intermediate-task learning is a better paradigm to introduce auxiliary tasks to AMR parsing, compared to multitask learning. From an empirical perspective, we propose a principled method to involve auxiliary tasks to boost AMR parsing. Extensive experiments show that our method achieves new state-of-the-art performance on different benchmarks especially in topology-related scores.

Peiyi Wang, Yifan Song, Tianyu Liu et al. · pku

Continual relation extraction (CRE) requires the model to continually learn new relations from class-incremental data streams. In this paper, we propose a Frustratingly easy but Effective Approach (FEA) method with two learning stages for CRE: 1) Fast Adaption (FA) warms up the model with only new data. 2) Balanced Tuning (BT) finetunes the model on the balanced memory data. Despite its simplicity, FEA achieves comparable (on TACRED or superior (on FewRel) performance compared with the state-of-the-art baselines. With careful examinations, we find that the data imbalance between new and old relations leads to a skewed decision boundary in the head classifiers over the pretrained encoders, thus hurting the overall performance. In FEA, the FA stage unleashes the potential of memory data for the subsequent finetuning, while the BT stage helps establish a more balanced decision boundary. With a unified view, we find that two strong CRE baselines can be subsumed into the proposed training pipeline. The success of FEA also provides actionable insights and suggestions for future model designing in CRE.

Afra Amini, Tianyu Liu, Ryan Cotterell · eth-zurich

We introduce a novel dependency parser, the hexatagger, that constructs dependency trees by tagging the words in a sentence with elements from a finite set of possible tags. In contrast to many approaches to dependency parsing, our approach is fully parallelizable at training time, i.e., the structure-building actions needed to build a dependency parse can be predicted in parallel to each other. Additionally, exact decoding is linear in time and space complexity. Furthermore, we derive a probabilistic dependency parser that predicts hexatags using no more than a linear model with features from a pretrained language model, i.e., we forsake a bespoke architecture explicitly designed for the task. Despite the generality and simplicity of our approach, we achieve state-of-the-art performance of 96.4 LAS and 97.4 UAS on the Penn Treebank test set. Additionally, our parser's linear time complexity and parallelism significantly improve computational efficiency, with a roughly 10-times speed-up over previous state-of-the-art models during decoding.

Shoujie Tong, Qingxiu Dong, Damai Dai et al. · pku

Fine-tuning pretrained language models (PLMs) on downstream tasks has become common practice in natural language processing. However, most of the PLMs are vulnerable, e.g., they are brittle under adversarial attacks or imbalanced data, which hinders the application of the PLMs on some downstream tasks, especially in safe-critical scenarios. In this paper, we propose a simple yet effective fine-tuning method called Match-Tuning to force the PLMs to be more robust. For each instance in a batch, we involve other instances in the same batch to interact with it. To be specific, regarding the instances with other labels as a perturbation, Match-Tuning makes the model more robust to noise at the beginning of training. While nearing the end, Match-Tuning focuses more on performing an interpolation among the instances with the same label for better generalization. Extensive experiments on various tasks in GLUE benchmark show that Match-Tuning consistently outperforms the vanilla fine-tuning by $1.64$ scores. Moreover, Match-Tuning exhibits remarkable robustness to adversarial attacks and data imbalance.

Yuchen Eleanor Jiang, Tianyu Liu, Shuming Ma et al. · eth-zurich

Machine translation (MT) has almost achieved human parity at sentence-level translation. In response, the MT community has, in part, shifted its focus to document-level translation. However, the development of document-level MT systems is hampered by the lack of parallel document corpora. This paper describes BWB, a large parallel corpus first introduced in Jiang et al. (2022), along with an annotated test set. The BWB corpus consists of Chinese novels translated by experts into English, and the annotated test set is designed to probe the ability of machine translation systems to model various discourse phenomena. Our resource is freely available, and we hope it will serve as a guide and inspiration for more work in document-level machine translation.

Tianyu Liu, Yuchen Jiang, Nicholas Monath et al.

Recent years have seen a paradigm shift in NLP towards using pretrained language models ({PLM}) for a wide range of tasks. However, there are many difficult design decisions to represent structures (e.g. tagged text, coreference chains) in a way such that they can be captured by PLMs. Prior work on structured prediction with PLMs typically flattens the structured output into a sequence, which limits the quality of structural information being learned and leads to inferior performance compared to classic discriminative models. In this work, we describe an approach to model structures as sequences of actions in an autoregressive manner with PLMs, allowing in-structure dependencies to be learned without any loss. Our approach achieves the new state-of-the-art on all the structured prediction tasks we looked at, namely, named entity recognition, end-to-end relation extraction, and coreference resolution.

Ryan Cotterell, Anej Svete, Clara Meister et al. · allen-ai, eth-zurich

Large language models have become one of the most commonly deployed NLP inventions. In the past half-decade, their integration into core natural language processing tools has dramatically increased the performance of such tools, and they have entered the public discourse surrounding artificial intelligence. Consequently, it is important for both developers and researchers alike to understand the mathematical foundations of large language models, as well as how to implement them. These notes are the accompaniment to the theoretical portion of the ETH Zürich course on large language models, covering what constitutes a language model from a formal, theoretical perspective.

Haoran Meng, Zheng Xin, Tianyu Liu et al.

While interacting with chatbots, users may elicit multiple intents in a single dialogue utterance. Instead of training a dedicated multi-intent detection model, we propose DialogUSR, a dialogue utterance splitting and reformulation task that first splits multi-intent user query into several single-intent sub-queries and then recovers all the coreferred and omitted information in the sub-queries. DialogUSR can serve as a plug-in and domain-agnostic module that empowers the multi-intent detection for the deployed chatbots with minimal efforts. We collect a high-quality naturally occurring dataset that covers 23 domains with a multi-step crowd-souring procedure. To benchmark the proposed dataset, we propose multiple action-based generative models that involve end-to-end and two-stage training, and conduct in-depth analyses on the pros and cons of the proposed baselines.

Tianyu Liu, Yuchen Eleanor Jiang, Ryan Cotterell et al.

Many natural language processing tasks, e.g., coreference resolution and semantic role labeling, require selecting text spans and making decisions about them. A typical approach to such tasks is to score all possible spans and greedily select spans for task-specific downstream processing. This approach, however, does not incorporate any inductive bias about what sort of spans ought to be selected, e.g., that selected spans tend to be syntactic constituents. In this paper, we propose a novel grammar-based structured span selection model which learns to make use of the partial span-level annotation provided for such problems. Compared to previous approaches, our approach gets rid of the heuristic greedy span selection scheme, allowing us to model the downstream task on an optimal set of spans. We evaluate our model on two popular span prediction tasks: coreference resolution and semantic role labeling. We show empirical improvements on both.

Tianyu Liu, Jie Lu, Zheng Yan et al.

As a powerful Bayesian non-parameterized algorithm, the Gaussian process (GP) has performed a significant role in Bayesian optimization and signal processing. GPs have also advanced online decision-making systems because their posterior distribution has a closed-form solution. However, its training and inference process requires all historic data to be stored and the GP model to be trained from scratch. For those reasons, several online GP algorithms, such as O-SGPR and O-SVGP, have been specifically designed for streaming settings. In this paper, we present a new theoretical framework for online GPs based on the online probably approximately correct (PAC) Bayes theory. The framework offers both a guarantee of generalized performance and good accuracy. Instead of minimizing the marginal likelihood, our algorithm optimizes both the empirical risk function and a regularization item, which is in proportion to the divergence between the prior distribution and posterior distribution of parameters. In addition to its theoretical appeal, the algorithm performs well empirically on several regression datasets. Compared to other online GP algorithms, ours yields a generalization guarantee and very competitive accuracy.

Xin Zheng, Tianyu Liu, Haoran Meng et al.

Harvesting question-answer (QA) pairs from customer service chatlog in the wild is an efficient way to enrich the knowledge base for customer service chatbots in the cold start or continuous integration scenarios. Prior work attempts to obtain 1-to-1 QA pairs from growing customer service chatlog, which fails to integrate the incomplete utterances from the dialog context for composite QA retrieval. In this paper, we propose N-to-N QA extraction task in which the derived questions and corresponding answers might be separated across different utterances. We introduce a suite of generative/discriminative tagging based methods with end-to-end and two-stage variants that perform well on 5 customer service datasets and for the first time setup a benchmark for N-to-N DialogQAE with utterance and session level evaluation metrics. With a deep dive into extracted QA pairs, we find that the relations between and inside the QA pairs can be indicators to analyze the dialogue structure, e.g. information seeking, clarification, barge-in and elaboration. We also show that the proposed models can adapt to different domains and languages, and reduce the labor cost of knowledge accumulation in the real-world product dialogue platform.

Hua Liang, Tianyu Liu, Peiyi Wang et al.

In modern sales applications, automatic script extraction and management greatly decrease the need for human labor to collect the winning sales scripts, which largely boost the success rate for sales and can be shared across the sales teams. In this work, we present the SmartSales system to serve both the sales representatives and managers to attain the sales insights from the large-scale sales chatlog. SmartSales consists of three modules: 1) Customer frequently asked questions (FAQ) extraction aims to enrich the FAQ knowledge base by harvesting high quality customer question-answer pairs from the chatlog. 2) Customer objection response assists the salespeople to figure out the typical customer objections and corresponding winning sales scripts, as well as search for proper sales responses for a certain customer objection. 3) Sales manager dashboard helps sales managers to monitor whether a specific sales representative or team follows the sales standard operating procedures (SOP). The proposed prototype system is empowered by the state-of-the-art conversational intelligence techniques and has been running on the Tencent Cloud to serve the sales teams from several different areas.

Jiaheng Liu, Dawei Zhu, Zhiqi Bai et al. · pku

Efficient processing of long contexts has been a persistent pursuit in Natural Language Processing. With the growing number of long documents, dialogues, and other textual data, it is important to develop Long Context Language Models (LCLMs) that can process and analyze extensive inputs in an effective and efficient way. In this paper, we present a comprehensive survey on recent advances in long-context modeling for large language models. Our survey is structured around three key aspects: how to obtain effective and efficient LCLMs, how to train and deploy LCLMs efficiently, and how to evaluate and analyze LCLMs comprehensively. For the first aspect, we discuss data strategies, architectural designs, and workflow approaches oriented with long context processing. For the second aspect, we provide a detailed examination of the infrastructure required for LCLM training and inference. For the third aspect, we present evaluation paradigms for long-context comprehension and long-form generation, as well as behavioral analysis and mechanism interpretability of LCLMs. Beyond these three key aspects, we thoroughly explore the diverse application scenarios where existing LCLMs have been deployed and outline promising future development directions. This survey provides an up-to-date review of the literature on long-context LLMs, which we wish to serve as a valuable resource for both researchers and engineers. An associated GitHub repository collecting the latest papers and repos is available at: \href{https://github.com/LCLM-Horizon/A-Comprehensive-Survey-For-Long-Context-Language-Modeling}{\color[RGB]{175,36,67}{LCLM-Horizon}}.

Rui-Jie Zhu, Tianhao Peng, Tianhao Cheng et al.

Large Language Models (LLMs) have demonstrated impressive reasoning capabilities, especially when guided by explicit chain-of-thought (CoT) reasoning that verbalizes intermediate steps. While CoT improves both interpretability and accuracy, its dependence on natural language reasoning limits the model's expressive bandwidth. Latent reasoning tackles this bottleneck by performing multi-step inference entirely in the model's continuous hidden state, eliminating token-level supervision. To advance latent reasoning research, this survey provides a comprehensive overview of the emerging field of latent reasoning. We begin by examining the foundational role of neural network layers as the computational substrate for reasoning, highlighting how hierarchical representations support complex transformations. Next, we explore diverse latent reasoning methodologies, including activation-based recurrence, hidden state propagation, and fine-tuning strategies that compress or internalize explicit reasoning traces. Finally, we discuss advanced paradigms such as infinite-depth latent reasoning via masked diffusion models, which enable globally consistent and reversible reasoning processes. By unifying these perspectives, we aim to clarify the conceptual landscape of latent reasoning and chart future directions for research at the frontier of LLM cognition. An associated GitHub repository collecting the latest papers and repos is available at: https://github.com/multimodal-art-projection/LatentCoT-Horizon/.

Peiyi Wang, Lei Li, Liang Chen et al.

In this paper, we uncover a systematic bias in the evaluation paradigm of adopting large language models~(LLMs), e.g., GPT-4, as a referee to score and compare the quality of responses generated by candidate models. We find that the quality ranking of candidate responses can be easily hacked by simply altering their order of appearance in the context. This manipulation allows us to skew the evaluation result, making one model appear considerably superior to the other, e.g., Vicuna-13B could beat ChatGPT on 66 over 80 tested queries with ChatGPT as an evaluator. To address this issue, we propose a calibration framework with three simple yet effective strategies: 1) Multiple Evidence Calibration, which requires the evaluator model to generate multiple evaluation evidence before assigning ratings; 2) Balanced Position Calibration, which aggregates results across various orders to determine the final score; 3) Human-in-the-Loop Calibration, which introduces a balanced position diversity entropy to measure the difficulty of each example and seeks human assistance when needed. We also manually annotate the "win/tie/lose" outcomes of responses from ChatGPT and Vicuna-13B in the Vicuna Benchmark's question prompt, and extensive experiments demonstrate that our approach successfully mitigates evaluation bias, resulting in closer alignment with human judgments. We release our code and human annotation at \url{https://github.com/i-Eval/FairEval} to facilitate future research.

Peiyi Wang, Runxin Xu, Tianyu Liu et al.

Few-Shot Sequence Labeling (FSSL) is a canonical paradigm for the tagging models, e.g., named entity recognition and slot filling, to generalize on an emerging, resource-scarce domain. Recently, the metric-based meta-learning framework has been recognized as a promising approach for FSSL. However, most prior works assign a label to each token based on the token-level similarities, which ignores the integrality of named entities or slots. To this end, in this paper, we propose ESD, an Enhanced Span-based Decomposition method for FSSL. ESD formulates FSSL as a span-level matching problem between test query and supporting instances. Specifically, ESD decomposes the span matching problem into a series of span-level procedures, mainly including enhanced span representation, class prototype aggregation and span conflicts resolution. Extensive experiments show that ESD achieves the new state-of-the-art results on two popular FSSL benchmarks, FewNERD and SNIPS, and is proven to be more robust in the nested and noisy tagging scenarios. Our code is available at https://github.com/Wangpeiyi9979/ESD.

Tim Vieira, Tianyu Liu, Clemente Pasti et al.

Modern language models represent probability distributions over character strings as distributions over (shorter) token strings derived via a deterministic tokenizer, such as byte-pair encoding. While this approach is highly effective at scaling up language models to large corpora, its current incarnations have a concerning property: the model assigns nonzero probability mass to an exponential number of $\it{noncanonical}$ token encodings of each character string -- these are token strings that decode to valid character strings but are impossible under the deterministic tokenizer (i.e., they will never be seen in any training corpus, no matter how large). This misallocation is both erroneous, as noncanonical strings never appear in training data, and wasteful, diverting probability mass away from plausible outputs. These are avoidable mistakes! In this work, we propose methods to enforce canonicality in token-level language models, ensuring that only canonical token strings are assigned positive probability. We present two approaches: (1) canonicality by conditioning, leveraging test-time inference strategies without additional training, and (2) canonicality by construction, a model parameterization that guarantees canonical outputs but requires training. We demonstrate that fixing canonicality mistakes improves the likelihood of held-out data for several models and corpora.

Ke Zhu, Minghao Fu, Jie Shao et al.

Long-tailed object detection faces great challenges because of its extremely imbalanced class distribution. Recent methods mainly focus on the classification bias and its loss function design, while ignoring the subtle influence of the regression branch. This paper shows that the regression bias exists and does adversely and seriously impact the detection accuracy. While existing methods fail to handle the regression bias, the class-specific regression head for rare classes is hypothesized to be the main cause of it in this paper. As a result, three kinds of viable solutions to cater for the rare categories are proposed, including adding a class-agnostic branch, clustering heads and merging heads. The proposed methods brings in consistent and significant improvements over existing long-tailed detection methods, especially in rare and common classes. The proposed method achieves state-of-the-art performance in the large vocabulary LVIS dataset with different backbones and architectures. It generalizes well to more difficult evaluation metrics, relatively balanced datasets, and the mask branch. This is the first attempt to reveal and explore rectifying of the regression bias in long-tailed object detection.

Feifan Song, Shaohang Wei, Bofei Gao et al. · pku

Large reasoning models (LRMs) boosted by Reinforcement Learning from Verifier Reward (RLVR) have shown great power in problem solving, yet they often cause overthinking: excessive, meandering reasoning that inflates computational cost. Prior designs of penalization in RLVR manage to reduce token consumption while often harming model performance, which arises from the oversimplicity of token-level supervision. In this paper, we argue that the granularity of supervision plays a crucial role in balancing efficiency and accuracy, and propose Group Relative Segment Penalization (GRSP), a step-level method to regularize reasoning. Since preliminary analyses show that reasoning segments are strongly correlated with token consumption and model performance, we design a length-aware weighting mechanism across segment clusters. Extensive experiments demonstrate that GRSP achieves superior token efficiency without heavily compromising accuracy, especially the advantages with harder problems. Moreover, GRSP stabilizes RL training and scales effectively across model sizes.

M-A-P Team, Xinrun Du, Yifan Yao et al.

Large language models (LLMs) have demonstrated remarkable proficiency in mainstream academic disciplines such as mathematics, physics, and computer science. However, human knowledge encompasses over 200 specialized disciplines, far exceeding the scope of existing benchmarks. The capabilities of LLMs in many of these specialized fields-particularly in light industry, agriculture, and service-oriented disciplines-remain inadequately evaluated. To address this gap, we present SuperGPQA, a comprehensive benchmark that evaluates graduate-level knowledge and reasoning capabilities across 285 disciplines. Our benchmark employs a novel Human-LLM collaborative filtering mechanism to eliminate trivial or ambiguous questions through iterative refinement based on both LLM responses and expert feedback. Our experimental results reveal significant room for improvement in the performance of current state-of-the-art LLMs across diverse knowledge domains (e.g., the reasoning-focused model DeepSeek-R1 achieved the highest accuracy of 61.82% on SuperGPQA), highlighting the considerable gap between current model capabilities and artificial general intelligence. Additionally, we present comprehensive insights from our management of a large-scale annotation process, involving over 80 expert annotators and an interactive Human-LLM collaborative system, offering valuable methodological guidance for future research initiatives of comparable scope.

Tianyu Liu, Kevin Du, Mrinmaya Sachan et al.

One strength of modern language models is their ability to incorporate information from a user-input context when answering queries. However, they are not equally sensitive to the subtle changes to that context. To quantify this, Du et al. (2024) gives an information-theoretic metric to measure such sensitivity. Their metric, susceptibility, is defined as the degree to which contexts can influence a model's response to a query at a distributional level. However, exactly computing susceptibility is difficult and, thus, Du et al. (2024) falls back on a Monte Carlo approximation. Due to the large number of samples required, the Monte Carlo approximation is inefficient in practice. As a faster alternative, we propose Fisher susceptibility, an efficient method to estimate the susceptibility based on Fisher information. Empirically, we validate that Fisher susceptibility is comparable to Monte Carlo estimated susceptibility across a diverse set of query domains despite its being $70\times$ faster. Exploiting the improved efficiency, we apply Fisher susceptibility to analyze factors affecting the susceptibility of language models. We observe that larger models are as susceptible as smaller ones.

Naaman Tan, Josef Valvoda, Tianyu Liu et al.

The relationship between the quality of a string, as judged by a human reader, and its probability, $p(\boldsymbol{y})$ under a language model undergirds the development of better language models. For example, many popular algorithms for sampling from a language model have been conceived with the goal of manipulating $p(\boldsymbol{y})$ to place higher probability on strings that humans deem of high quality. In this article, we examine the probability--quality relationship in language models explicitly aligned to human preferences, e.g., through reinforcement learning through human feedback. We show that, when sampling corpora from an aligned language model, there exists a trade-off between the strings' average reward and average log-likelihood under the prior language model, i.e., the same model before alignment with human preferences. We provide a formal treatment of this phenomenon and demonstrate how a choice of sampling adaptor allows for a selection of how much likelihood we exchange for the reward.

Zefan Cai, Yichi Zhang, Bofei Gao et al.

In this study, we investigate whether attention-based information flow inside large language models (LLMs) is aggregated through noticeable patterns for long context processing. Our observations reveal that LLMs aggregate information through Pyramidal Information Funneling where attention is scattering widely in lower layers, progressively consolidating within specific contexts, and ultimately focusing on critical tokens (a.k.a massive activation or attention sink) in higher layers. Motivated by these insights, we developed PyramidKV, a novel and effective KV cache compression method. This approach dynamically adjusts the KV cache size across different layers, allocating more cache in lower layers and less in higher ones, diverging from traditional methods that maintain a uniform KV cache size. Our experimental evaluations, utilizing the LongBench benchmark, show that PyramidKV matches the performance of models with a full KV cache while retaining only 12% of the KV cache, thus significantly reducing memory usage. In scenarios emphasizing memory efficiency, where only 0.7% of the KV cache is maintained, PyramidKV surpasses other KV cache compression techniques, achieving up to a 20.5 absolute accuracy improvement on TREC dataset. In the Needle-in-a-Haystack experiment, PyramidKV outperforms competing methods in maintaining long-context comprehension in LLMs; notably, retaining just 128 KV cache entries enables the LLAMA-3-70B model to achieve 100.0 Acc. performance.

Tianyu Liu, Afra Amini, Mrinmaya Sachan et al.

Tasks that model the relation between pairs of tokens in a string are a vital part of understanding natural language. Such tasks, in general, require exhaustive pair-wise comparisons of tokens, thus having a quadratic runtime complexity in the length of the string. We show that these exhaustive comparisons can be avoided, and, moreover, the complexity of such tasks can be reduced to linear by casting the relation between tokens as a partial order over the string. Our method predicts real numbers for each token in a string in parallel and sorts the tokens accordingly, resulting in total orders of the tokens in the string. Each total order implies a set of arcs oriented from smaller to greater tokens, sorted by their predicted numbers. The intersection of total orders results in a partial order over the set of tokens in the string, which is then decoded into a directed graph representing the desired linguistic structure. Our experiments on dependency parsing and coreference resolution show that our method achieves state-of-the-art or comparable performance. Moreover, the linear complexity and parallelism of our method double the speed of graph-based coreference resolution models, and bring a 10-times speed-up over graph-based dependency parsers.

Shoujie Tong, Heming Xia, Damai Dai et al.

Pretrained language models have achieved remarkable success in natural language understanding. However, fine-tuning pretrained models on limited training data tends to overfit and thus diminish performance. This paper presents Bi-Drop, a fine-tuning strategy that selectively updates model parameters using gradients from various sub-nets dynamically generated by dropout. The sub-net estimation of Bi-Drop is performed in an in-batch manner, so it overcomes the problem of hysteresis in sub-net updating, which is possessed by previous methods that perform asynchronous sub-net estimation. Also, Bi-Drop needs only one mini-batch to estimate the sub-net so it achieves higher utility of training data. Experiments on the GLUE benchmark demonstrate that Bi-Drop consistently outperforms previous fine-tuning methods. Furthermore, empirical results also show that Bi-Drop exhibits excellent generalization ability and robustness for domain transfer, data imbalance, and low-resource scenarios.

Zefan Cai, Xin Zheng, Tianyu Liu et al.

In the constant updates of the product dialogue systems, we need to retrain the natural language understanding (NLU) model as new data from the real users would be merged into the existent data accumulated in the last updates. Within the newly added data, new intents would emerge and might have semantic entanglement with the existing intents, e.g. new intents that are semantically too specific or generic are actually subset or superset of some existing intents in the semantic space, thus impairing the robustness of the NLU model. As the first attempt to solve this problem, we setup a new benchmark consisting of 4 Dialogue Version Control dataSets (DialogVCS). We formulate the intent detection with imperfect data in the system update as a multi-label classification task with positive but unlabeled intents, which asks the models to recognize all the proper intents, including the ones with semantic entanglement, in the inference. We also propose comprehensive baseline models and conduct in-depth analyses for the benchmark, showing that the semantically entangled intents can be effectively recognized with an automatic workflow.

Yifan Song, Peiyi Wang, Dawei Zhu et al.

Continual learning (CL) aims to constantly learn new knowledge over time while avoiding catastrophic forgetting on old tasks. In this work, we focus on continual text classification under the class-incremental setting. Recent CL studies find that the representations learned in one task may not be effective for other tasks, namely representation bias problem. For the first time we formally analyze representation bias from an information bottleneck perspective and suggest that exploiting representations with more class-relevant information could alleviate the bias. To this end, we propose a novel replay-based continual text classification method, RepCL. Our approach utilizes contrastive and generative representation learning objectives to capture more class-relevant features. In addition, RepCL introduces an adversarial replay strategy to alleviate the overfitting problem of replay. Experiments demonstrate that RepCL effectively alleviates forgetting and achieves state-of-the-art performance on three text classification tasks.

Wei Huang, Chunrui Liu, Yilan Chen et al.

PAC-Bayesian is an analysis framework where the training error can be expressed as the weighted average of the hypotheses in the posterior distribution whilst incorporating the prior knowledge. In addition to being a pure generalization bound analysis tool, PAC-Bayesian bound can also be incorporated into an objective function to train a probabilistic neural network, making them a powerful and relevant framework that can numerically provide a tight generalization bound for supervised learning. For simplicity, we call probabilistic neural network learned using training objectives derived from PAC-Bayesian bounds as {\it PAC-Bayesian learning}. Despite their empirical success, the theoretical analysis of PAC-Bayesian learning for neural networks is rarely explored. This paper proposes a new class of convergence and generalization analysis for PAC-Bayes learning when it is used to train the over-parameterized neural networks by the gradient descent method. For a wide probabilistic neural network, we show that when PAC-Bayes learning is applied, the convergence result corresponds to solving a kernel ridge regression when the probabilistic neural tangent kernel (PNTK) is used as its kernel. Based on this finding, we further characterize the uniform PAC-Bayesian generalization bound which improves over the Rademacher complexity-based bound for non-probabilistic neural network. Finally, drawing the insight from our theoretical results, we propose a proxy measure for efficient hyperparameters selection, which is proven to be time-saving.

Peiyi Wang, Liang Chen, Tianyu Liu et al.

Abstract Meaning Representation (AMR) parsing aims to translate sentences to semantic representation with a hierarchical structure, and is recently empowered by pretrained sequence-to-sequence models. However, there exists a gap between their flat training objective (i.e., equally treats all output tokens) and the hierarchical AMR structure, which limits the model generalization. To bridge this gap, we propose a Hierarchical Curriculum Learning (HCL) framework with Structure-level (SC) and Instance-level Curricula (IC). SC switches progressively from core to detail AMR semantic elements while IC transits from structure-simple to -complex AMR instances during training. Through these two warming-up processes, HCL reduces the difficulty of learning complex structures, thus the flat model can better adapt to the AMR hierarchy. Extensive experiments on AMR2.0, AMR3.0, structure-complex and out-of-distribution situations verify the effectiveness of HCL.

Peiyi Wang, Runxin Xu, Tianyu Liu et al.

Few-Shot Event Classification (FSEC) aims at developing a model for event prediction, which can generalize to new event types with a limited number of annotated data. Existing FSEC studies have achieved high accuracy on different benchmarks. However, we find they suffer from trigger biases that signify the statistical homogeneity between some trigger words and target event types, which we summarize as trigger overlapping and trigger separability. The biases can result in context-bypassing problem, i.e., correct classifications can be gained by looking at only the trigger words while ignoring the entire context. Therefore, existing models can be weak in generalizing to unseen data in real scenarios. To further uncover the trigger biases and assess the generalization ability of the models, we propose two new sampling methods, Trigger-Uniform Sampling (TUS) and COnfusion Sampling (COS), for the meta tasks construction during evaluation. Besides, to cope with the context-bypassing problem in FSEC models, we introduce adversarial training and trigger reconstruction techniques. Experiments show these techniques help not only improve the performance, but also enhance the generalization ability of models.

Peiyi Wang, Tianyu Liu, Damai Dai et al.

Aspect Sentiment Triplet Extraction (ASTE) aims to recognize targets, their sentiment polarities and opinions explaining the sentiment from a sentence. ASTE could be naturally divided into 3 atom subtasks, namely target detection, opinion detection and sentiment classification. We argue that the proper subtask combination, compositional feature extraction for target-opinion pairs, and interaction between subtasks would be the key to success. Prior work, however, may fail on `one-to-many' or `many-to-one' situations or derive non-existent sentiment triplets due to defective subtask formulation, sub-optimal feature representation or the lack of subtask interaction. In this paper, we divide ASTE into target-opinion joint detection and sentiment classification subtasks, which is in line with human cognition, and correspondingly utilize sequence encoder and table encoder to handle them. Table encoder extracts sentiment at token-pair level, so that the compositional feature between targets and opinions can be easily captured. To establish explicit interaction between subtasks, we utilize the table representation to guide the sequence encoding, and inject the sequence features back into the table encoder. Experiments show that our model outperforms state-of-the-art methods on six popular ASTE datasets.

Qingxiu Dong, Ziwei Qin, Heming Xia et al.

It is a common practice for recent works in vision language cross-modal reasoning to adopt a binary or multi-choice classification formulation taking as input a set of source image(s) and textual query. In this work, we take a sober look at such an unconditional formulation in the sense that no prior knowledge is specified with respect to the source image(s). Inspired by the designs of both visual commonsense reasoning and natural language inference tasks, we propose a new task termed Premise-based Multi-modal Reasoning(PMR) where a textual premise is the background presumption on each source image. The PMR dataset contains 15,360 manually annotated samples which are created by a multi-phase crowd-sourcing process. With selected high-quality movie screenshots and human-curated premise templates from 6 pre-defined categories, we ask crowd-source workers to write one true hypothesis and three distractors (4 choices) given the premise and image through a cross-check procedure. Besides, we generate adversarial samples to alleviate the annotation artifacts and double the size of PMR. We benchmark various state-of-the-art (pretrained) multi-modal inference models on PMR and conduct comprehensive experimental analyses to showcase the utility of our dataset.

Tianyu Liu, Lingyu Zhang

The manpower scheduling problem is a kind of critical combinational optimization problem. Researching solutions to scheduling problems can improve the efficiency of companies, hospitals, and other work units. This paper proposes a new model combined with deep learning to solve the multi-shift manpower scheduling problem based on the existing research. This model first solves the objective function's optimized value according to the current constraints to find the plan of employee arrangement initially. It will then use the scheduling table generation algorithm to obtain the scheduling result in a short time. Moreover, the most prominent feature we propose is that we will use the neural network training method based on the time series to solve long-term and long-period scheduling tasks and obtain manpower arrangement. The selection criteria of the neural network and the training process are also described in this paper. We demonstrate that our model can make a precise forecast based on the improvement of neural networks. This paper also discusses the challenges in the neural network training process and obtains enlightening results after getting the arrangement plan. Our research shows that neural networks and deep learning strategies have the potential to solve similar problems effectively.

Tianyu Liu, Yizhe Zhang, Chris Brockett et al.

Large pretrained generative models like GPT-3 often suffer from hallucinating non-existent or incorrect content, which undermines their potential merits in real applications. Existing work usually attempts to detect these hallucinations based on a corresponding oracle reference at a sentence or document level. However ground-truth references may not be readily available for many free-form text generation applications, and sentence- or document-level detection may fail to provide the fine-grained signals that would prevent fallacious content in real time. As a first step to addressing these issues, we propose a novel token-level, reference-free hallucination detection task and an associated annotated dataset named HaDes (HAllucination DEtection dataSet). To create this dataset, we first perturb a large number of text segments extracted from English language Wikipedia, and then verify these with crowd-sourced annotations. To mitigate label imbalance during annotation, we utilize an iterative model-in-loop strategy. We conduct comprehensive data analyses and create multiple baseline models.

Tianyu Liu, Xin Zheng, Baobao Chang et al.

In open domain table-to-text generation, we notice that the unfaithful generation usually contains hallucinated content which can not be aligned to any input table record. We thus try to evaluate the generation faithfulness with two entity-centric metrics: table record coverage and the ratio of hallucinated entities in text, both of which are shown to have strong agreement with human judgements. Then based on these metrics, we quantitatively analyze the correlation between training data quality and generation fidelity which indicates the potential usage of entity information in faithful generation. Motivated by these findings, we propose two methods for faithful generation: 1) augmented training by incorporating the auxiliary entity information, including both an augmented plan-based model and an unsupervised model and 2) training instance selection based on faithfulness ranking. We show these approaches improve generation fidelity in both full dataset setting and few shot learning settings by both automatic and human evaluations.

Tianyu Liu, Jie Lu, Zheng Yan et al.

By leveraging experience from previous tasks, meta-learning algorithms can achieve effective fast adaptation ability when encountering new tasks. However it is unclear how the generalization property applies to new tasks. Probably approximately correct (PAC) Bayes bound theory provides a theoretical framework to analyze the generalization performance for meta-learning. We derive three novel generalisation error bounds for meta-learning based on PAC-Bayes relative entropy bound. Furthermore, using the empirical risk minimization (ERM) method, a PAC-Bayes bound for meta-learning with data-dependent prior is developed. Experiments illustrate that the proposed three PAC-Bayes bounds for meta-learning guarantee a competitive generalization performance guarantee, and the extended PAC-Bayes bound with data-dependent prior can achieve rapid convergence ability.

Zona Kostic, Jared Jessup, Jeffrey Baglioni et al.

An innumerable number of individual choices go into discovering a new book. There are unmistakably two groups of booklovers: those who like to search online, follow other people's latest readings, or simply react to a system's recommendations; and those who love to wander between library stacks, lose themselves behind bookstore shelves, or simply hide behind piles of (un)organized books. Depending on which group a person may fall into, there are two distinct and corresponding mediums that inform his or her choices: digital, that provides efficient retrieval of information online, and physical, a more tactile pursuit that leads to unexpected discoveries and promotes serendipity. How could we possibly bridge the gap between these seemingly disparate mediums into an integrated system that can amplify the benefits they both offer? In this paper, we present the BookVIS application, which uses book-related data and generates personalized visualizations to follow users in their quest for a new book. In this new redesigned version, the app brings associative visual connections to support intuitive exploration of easily retrieved digital information and its relationship with the physical book in hand. BookVIS keeps track of the user's reading preferences and generates a dataSelfie as an individual snapshot of a personal taste that grows over time. Usability testing has also been conducted and has demonstrated the app's ability to identify distinguishable patterns in readers' tastes that could be further used to communicate personal preferences in new "shelf-browsing" iterations. By efficiently supplementing the user's cognitive information needs while still supporting the spontaneity and enjoyment of the book browsing experience, BookVIS bridges the gap between real and online realms, and maximizes the engagement of personalized mobile visual clues.

Tianyu Liu, Xin Zheng, Xiaoan Ding et al.

The prior work on natural language inference (NLI) debiasing mainly targets at one or few known biases while not necessarily making the models more robust. In this paper, we focus on the model-agnostic debiasing strategies and explore how to (or is it possible to) make the NLI models robust to multiple distinct adversarial attacks while keeping or even strengthening the models' generalization power. We firstly benchmark prevailing neural NLI models including pretrained ones on various adversarial datasets. We then try to combat distinct known biases by modifying a mixture of experts (MoE) ensemble method and show that it's nontrivial to mitigate multiple NLI biases at the same time, and that model-level ensemble method outperforms MoE ensemble method. We also perform data augmentation including text swap, word substitution and paraphrase and prove its efficiency in combating various (though not all) adversarial attacks at the same time. Finally, we investigate several methods to merge heterogeneous training data (1.35M) and perform model ensembling, which are straightforward but effective to strengthen NLI models.

Tianyu Liu, Xin Zheng, Baobao Chang et al.

Many recent studies have shown that for models trained on datasets for natural language inference (NLI), it is possible to make correct predictions by merely looking at the hypothesis while completely ignoring the premise. In this work, we manage to derive adversarial examples in terms of the hypothesis-only bias and explore eligible ways to mitigate such bias. Specifically, we extract various phrases from the hypotheses (artificial patterns) in the training sets, and show that they have been strong indicators to the specific labels. We then figure out `hard' and `easy' instances from the original test sets whose labels are opposite to or consistent with those indications. We also set up baselines including both pretrained models (BERT, RoBERTa, XLNet) and competitive non-pretrained models (InferSent, DAM, ESIM). Apart from the benchmark and baselines, we also investigate two debiasing approaches which exploit the artificial pattern modeling to mitigate such hypothesis-only bias: down-sampling and adversarial training. We believe those methods can be treated as competitive baselines in NLI debiasing tasks.

Lei Sha, Lili Mou, Tianyu Liu et al.

Generating texts from structured data (e.g., a table) is important for various natural language processing tasks such as question answering and dialog systems. In recent studies, researchers use neural language models and encoder-decoder frameworks for table-to-text generation. However, these neural network-based approaches do not model the order of contents during text generation. When a human writes a summary based on a given table, he or she would probably consider the content order before wording. In a biography, for example, the nationality of a person is typically mentioned before occupation in a biography. In this paper, we propose an order-planning text generation model to capture the relationship between different fields and use such relationship to make the generated text more fluent and smooth. We conducted experiments on the WikiBio dataset and achieve significantly higher performance than previous methods in terms of BLEU, ROUGE, and NIST scores.