Junlin Wang

Margarita Geleta, Jiacen Xu, Manikanta Loya et al.

Although the prevention of AI vulnerabilities is critical to preserve the safety and privacy of users and businesses, educational tools for robust AI are still underdeveloped worldwide. We present the design, implementation, and assessment of Maestro. Maestro is an effective open-source game-based platform that contributes to the advancement of robust AI education. Maestro provides goal-based scenarios where college students are exposed to challenging life-inspired assignments in a competitive programming environment. We assessed Maestro's influence on students' engagement, motivation, and learning success in robust AI. This work also provides insights into the design features of online learning tools that promote active learning opportunities in the robust AI domain. We analyzed the reflection responses (measured with Likert scales) of 147 undergraduate students using Maestro in two quarterly college courses in AI. According to the results, students who felt the acquisition of new skills in robust AI tended to appreciate highly Maestro and scored highly on material consolidation, curiosity, and mastery in robust AI. Moreover, the leaderboard, our key gamification element in Maestro, has effectively contributed to students' engagement and learning. Results also indicate that Maestro can be effectively adapted to any course length and depth without losing its educational quality.

Junlin Wang

Learning visuomotor policies via behavior cloning typically involves mimicking expert demonstrations collected by human operators. However, natural human demonstrations inherently contain high-frequency noise, such as intermittent jerks, pauses, and action jitter. Training policies to directly imitate these raw trajectories inevitably causes the model to inherit these suboptimal behaviors. This pathology is particularly pronounced in diffusion-based policies, where iterative denoising steps can inadvertently amplify high-frequency artifacts at the expense of meaningful fine-grained details. To address these limitations, we present a novel frequency-based algorithm that enables implicit spectral maneuvering and smooth action generation. Our method, Frequency Guidance Operator (FGO), steers the generation process of diffusion polices by progressively driving the noisy samples through intermediate sub-frequency manifolds with expanding spectral bands. Validated on 15 robotic manipulation tasks from 5 benchmarks, FGO achieves superior performance in enhancing action smoothness and temporal consistency while preserving the details necessary for successful task execution. Project website: https://henrywjl.github.io/frequency-guidance-operator/

Junlin Wang, Federico Bianchi, Shang Zhu et al.

Modern AI benchmarks operate at a complexity that outpaces traditional verification methods. Tasks authored by domain experts often contain implicit assumptions, incomplete environment specifications, and brittle evaluation logic that human annotation cannot reliably catch. We introduce Auto Benchmark Audit (ABA), an agentic framework that systematically audits individual benchmark tasks, uncovering issues such as hidden environment dependencies, specification gaps, and limited grading logic. We run ABA on a collection of frontier LLM benchmarks and previous NeurIPS publications, totaling 168 benchmarks across nine domains. Across this corpus, ABA identifies critical issues including ambiguous task design, execution environment conflicts, and incorrect ground truths in over 25.7% of the evaluated tasks. The precision of these automated audits is validated by expert review and independent third-party reports such as upstream PRs. Crucially, we demonstrate that these problematic tasks severely distorts capability assessments for agents and LLMs: filtering out these tasks with issues shifts model rankings and increases average performance on SWE-bench Verified and Terminal-Bench 2 by 9.9% and 9.6%, respectively. We release the agentic tool and all task annotations to support the future development of frontier benchmarks.

Hongyang Yang, Boyu Zhang, Neng Wang et al.

As financial institutions and professionals increasingly incorporate Large Language Models (LLMs) into their workflows, substantial barriers, including proprietary data and specialized knowledge, persist between the finance sector and the AI community. These challenges impede the AI community's ability to enhance financial tasks effectively. Acknowledging financial analysis's critical role, we aim to devise financial-specialized LLM-based toolchains and democratize access to them through open-source initiatives, promoting wider AI adoption in financial decision-making. In this paper, we introduce FinRobot, a novel open-source AI agent platform supporting multiple financially specialized AI agents, each powered by LLM. Specifically, the platform consists of four major layers: 1) the Financial AI Agents layer that formulates Financial Chain-of-Thought (CoT) by breaking sophisticated financial problems down into logical sequences; 2) the Financial LLM Algorithms layer dynamically configures appropriate model application strategies for specific tasks; 3) the LLMOps and DataOps layer produces accurate models by applying training/fine-tuning techniques and using task-relevant data; 4) the Multi-source LLM Foundation Models layer that integrates various LLMs and enables the above layers to access them directly. Finally, FinRobot provides hands-on for both professional-grade analysts and laypersons to utilize powerful AI techniques for advanced financial analysis. We open-source FinRobot at \url{https://github.com/AI4Finance-Foundation/FinRobot}.

Hongwei Wu, Junlin Wang, Xin Wang et al.

It is a challenge to segment the location and size of rectal cancer tumours through deep learning. In this paper, in order to improve the ability of extracting suffi-cient feature information in rectal tumour segmentation, attention enlarged ConvNeXt UNet (AACN-UNet), is proposed. The network mainly includes two improvements: 1) the encoder stage of UNet is changed to ConvNeXt structure for encoding operation, which can not only integrate multi-scale semantic information on a large scale, but al-so reduce information loss and extract more feature information from CT images; 2) CBAM attention mechanism is added to improve the connection of each feature in channel and space, which is conducive to extracting the effective feature of the target and improving the segmentation accuracy.The experiment with UNet and its variant network shows that AACN-UNet is 0.9% ,1.1% and 1.4% higher than the current best results in P, F1 and Miou.Compared with the training time, the number of parameters in UNet network is less. This shows that our proposed AACN-UNet has achieved ex-cellent results in CT image segmentation of rectal cancer.

Junlin Wang, Roy Xie, Shang Zhu et al.

Building helpful and harmless large language models (LLMs) requires effective model alignment approach based on human instructions and feedback, which necessitates high-quality human-labeled data. Constructing such datasets is often expensive and hard to scale, and may face potential limitations on diversity and generalization. To address these challenges, we introduce Mixture of Agents Alignment (MoAA), that leverages the collective strengths of various language models to provide high-quality data for model alignment. By employing MoAA, we enhance both supervised fine-tuning and preference optimization, leading to improved performance compared to using a single model alone to generate alignment data (e.g. using GPT-4o alone). Evaluation results show that our approach can improve win rate of LLaMA-3.1-8B-Instruct from 19.5 to 48.3 on Arena-Hard and from 22.33 to 57.23 on AlpacaEval2, highlighting a promising direction for model alignment through this new scalable and diverse synthetic data recipe. Furthermore, we demonstrate that MoAA enables a self-improvement pipeline, where models finetuned on MoA-generated data surpass their own initial capabilities, providing evidence that our approach can push the frontier of open-source LLMs without reliance on stronger external supervision. Data and code will be released.

Fan Nie, Junlin Wang, Harper Hua et al.

Data science agents promise to accelerate discovery and insight-generation by turning data into executable analyses and findings. Yet existing data science benchmarks fall short due to fragmented evaluation interfaces that make cross-benchmark comparison difficult, narrow task coverage and a lack of rigorous data grounding. In particular, we show that a substantial portion of tasks in current benchmarks can be solved without using the actual data. To address these limitations, we introduce DSGym, a standardized framework for evaluating and training data science agents in self-contained execution environments. Unlike static benchmarks, DSGym provides a modular architecture that makes it easy to add tasks, agent scaffolds, and tools, positioning it as a live, extensible testbed. We curate DSGym-Tasks, a holistic task suite that standardizes and refines existing benchmarks via quality and shortcut solvability filtering. We further expand coverage with (1) DSBio: expert-derived bioinformatics tasks grounded in literature and (2) DSPredict: challenging prediction tasks spanning domains such as computer vision, molecular prediction, and single-cell perturbation. Beyond evaluation, DSGym enables agent training via execution-verified data synthesis pipeline. As a case study, we build a 2,000-example training set and trained a 4B model in DSGym that outperforms GPT-4o on standardized analysis benchmarks. Overall, DSGym enables rigorous end-to-end measurement of whether agents can plan, implement, and validate data analyses in realistic scientific context.

Ajay Vikram Periasami, Junlin Wang, Bhuwan Dhingra

Image-to-code generation tests whether a vision-language model (VLM) can recover the structure of an image enough to express it as executable code. Existing benchmarks either focus on narrow visual domains, depend on paired executable reference code, or rely on generic rubrics that miss domain-specific reconstruction errors. We introduce Vision2Code, a reference-code-free benchmark and evaluation framework for multi-domain image-to-code generation. Vision2Code contains 2,169 test examples from 15 source datasets that span charts and plots, geometry, graphs, scientific imagery, documents, and 3D spatial scenes. Models generate executable programs, which we render and score against the source image using a VLM rater with dataset-specific rubrics and deterministic guardrails for severe semantic failures. We report render-success diagnostics that separate code execution failures from reconstruction quality. Human validation shows that this evaluation protocol aligns better with human judgments than either a generic visual rubric or embedding-similarity baselines. Across nine open-weight and proprietary models, we find that image-to-code performance is domain-dependent: leading models perform well on regular chart- and graph-like visuals but remain weak on spatial scenes, chemistry, documents, and circuit-style diagrams. Finally, we show that evaluator-filtered model outputs can serve as training data to improve image-to-code capability, with Qwen3.5-9B improving from 1.60 to 1.86 on the benchmark without paired source programs. Vision2Code provides a reproducible testbed for measuring, diagnosing, and improving image-to-code generation. Our code and data are publicly available at https://image2code.github.io/vision2code/.

Junlin Wang, Zhiyun Lin

Learning effective visual representations for robotic manipulation remains a fundamental challenge due to the complex body dynamics involved in action execution. In this paper, we study how visual representations that carry body-relevant cues can enable efficient policy learning for downstream robotic manipulation tasks. We present $\textbf{I}$nter-token $\textbf{Con}$trast ($\textbf{ICon}$), a contrastive learning method applied to the token-level representations of Vision Transformers (ViTs). ICon enforces a separation in the feature space between agent-specific and environment-specific tokens, resulting in agent-centric visual representations that embed body-specific inductive biases. This framework can be seamlessly integrated into end-to-end policy learning by incorporating the contrastive loss as an auxiliary objective. Our experiments show that ICon not only improves policy performance across various manipulation tasks but also facilitates policy transfer across different robots. The project website: https://github.com/HenryWJL/icon

Junlin Wang, Tianyi Yang, Roy Xie et al.

With the proliferation of LLM-integrated applications such as GPT-s, millions are deployed, offering valuable services through proprietary instruction prompts. These systems, however, are prone to prompt extraction attacks through meticulously designed queries. To help mitigate this problem, we introduce the Raccoon benchmark which comprehensively evaluates a model's susceptibility to prompt extraction attacks. Our novel evaluation method assesses models under both defenseless and defended scenarios, employing a dual approach to evaluate the effectiveness of existing defenses and the resilience of the models. The benchmark encompasses 14 categories of prompt extraction attacks, with additional compounded attacks that closely mimic the strategies of potential attackers, alongside a diverse collection of defense templates. This array is, to our knowledge, the most extensive compilation of prompt theft attacks and defense mechanisms to date. Our findings highlight universal susceptibility to prompt theft in the absence of defenses, with OpenAI models demonstrating notable resilience when protected. This paper aims to establish a more systematic benchmark for assessing LLM robustness against prompt extraction attacks, offering insights into their causes and potential countermeasures. Resources of Raccoon are publicly available at https://github.com/M0gician/RaccoonBench.

Junlin Wang, Jue Wang, Ben Athiwaratkun et al.

Recent advances in large language models (LLMs) demonstrate substantial capabilities in natural language understanding and generation tasks. With the growing number of LLMs, how to harness the collective expertise of multiple LLMs is an exciting open direction. Toward this goal, we propose a new approach that leverages the collective strengths of multiple LLMs through a Mixture-of-Agents (MoA) methodology. In our approach, we construct a layered MoA architecture wherein each layer comprises multiple LLM agents. Each agent takes all the outputs from agents in the previous layer as auxiliary information in generating its response. MoA models achieves state-of-art performance on AlpacaEval 2.0, MT-Bench and FLASK, surpassing GPT-4 Omni. For example, our MoA using only open-source LLMs is the leader of AlpacaEval 2.0 by a substantial gap, achieving a score of 65.1% compared to 57.5% by GPT-4 Omni.

Junlin Wang, Zi Xu, Huiling Zhang

In this paper, we study second-order algorithms for the convex-concave minimax problem, which has attracted much attention in many fields such as machine learning in recent years. We propose a Lipschitz-free cubic regularization (LF-CR) algorithm for solving the convex-concave minimax optimization problem without knowing the Lipschitz constant. It can be shown that the iteration complexity of the LF-CR algorithm to obtain an $ε$-optimal solution with respect to the restricted primal-dual gap is upper bounded by $\mathcal{O}(ρ^{2/3}\|z_0-z^*\|^2ε^{-2/3})$ , where $z_0=(x_0,y_0)$ is a pair of initial points, $z^*=(x^*,y^*)$ is a pair of optimal solutions, and $ρ$ is the Lipschitz constant. We further propose a fully parameter-free cubic regularization (FF-CR) algorithm that does not require any parameters of the problem, including the Lipschitz constant and the upper bound of the distance from the initial point to the optimal solution. We also prove that the iteration complexity of the FF-CR algorithm to obtain an $ε$-optimal solution with respect to the gradient norm is upper bounded by $\mathcal{O}(ρ^{2/3}\|z_0-z^*\|^{4/3}ε^{-2/3}) $. Numerical experiments show the efficiency of both algorithms. To the best of our knowledge, the proposed FF-CR algorithm is a completely parameter-free second-order algorithm, and its iteration complexity is currently the best in terms of $ε$ under the termination criterion of the gradient norm.

Junlin Wang, Shang Zhu, Jon Saad-Falcon et al.

There is intense interest in investigating how inference time compute (ITC) (e.g. repeated sampling, refinements, etc) can improve large language model (LLM) capabilities. At the same time, recent breakthroughs in reasoning models, such as Deepseek-R1, unlock the opportunity for reinforcement learning to improve LLM reasoning skills. An in-depth understanding of how ITC interacts with reasoning across different models could provide important guidance on how to further advance the LLM frontier. This work conducts a comprehensive analysis of inference-time scaling methods for both reasoning and non-reasoning models on challenging reasoning tasks. Specifically, we focus our research on verifier-free inference time-scaling methods due to its generalizability without needing a reward model. We construct the Pareto frontier of quality and efficiency. We find that non-reasoning models, even with an extremely high inference budget, still fall substantially behind reasoning models. For reasoning models, majority voting proves to be a robust inference strategy, generally competitive or outperforming other more sophisticated ITC methods like best-of-N and sequential revisions, while the additional inference compute offers minimal improvements. We further perform in-depth analyses of the association of key response features (length and linguistic markers) with response quality, with which we can improve the existing ITC methods. We find that correct responses from reasoning models are typically shorter and have fewer hedging and thinking markers (but more discourse markers) than the incorrect responses.

Hongyi James Cai, Junlin Wang, Xiaoyin Chen et al.

Recent breakthroughs in large language models (LLMs) have effectively improved their reasoning abilities, particularly on mathematical and logical problems that have verifiable answers, through techniques such as supervised finetuning (SFT) and reinforcement learning (RL). Prior research indicates that RL effectively internalizes search strategies, enabling long chain-of-thought (CoT) reasoning, with backtracking emerging naturally as a learned capability. However, the precise benefits of backtracking, specifically, how significantly it contributes to reasoning improvements and the optimal extent of its use, remain poorly understood. In this work, we systematically investigate the dynamics between SFT and RL on eight reasoning tasks: Countdown, Sudoku, Arc 1D, Geometry, Color Cube Rotation, List Functions, Zebra Puzzles, and Self Reference. Our findings highlight that short CoT sequences used in SFT as a warm-up do have moderate contribution to RL training, compared with cold-start RL; however such contribution diminishes when tasks become increasingly difficult. Motivated by this observation, we construct synthetic datasets varying systematically in the number of backtracking steps and conduct controlled experiments to isolate the influence of either the correctness (content) or the structure (i.e., backtrack frequency). We find that (1) longer CoT with backtracks generally induce better and more stable RL training, (2) more challenging problems with larger search space tend to need higher numbers of backtracks during the SFT stage. Additionally, we demonstrate through experiments on distilled data that RL training is largely unaffected by the correctness of long CoT sequences, suggesting that RL prioritizes structural patterns over content correctness. Collectively, our results offer practical insights into designing optimal training strategies to effectively scale reasoning in LLMs.

Zhen Xu, Shang Zhu, Jue Wang et al.

We investigate the challenge of applying Large Language Models (LLMs) to long texts. We propose a theoretical framework that distinguishes the failure modes of long context tasks into three categories: cross-chunk dependence (task noise), confusion that grows with context size (model noise), and the imperfect integration of partial results (aggregator noise). Under this view, we analyze when it is effective to use multi-agent chunking, i.e., dividing a length sequence into smaller chunks and aggregating the processed results of each chunk. Our experiments on tasks such as retrieval, question answering, and summarization confirm both the theoretical analysis and the conditions that favor multi-agent chunking. By exploring superlinear model noise growth with input length, we also explain why, for large inputs, a weaker model configured with chunk-based processing can surpass a more advanced model like GPT4o applied in a single shot. Overall, we present a principled understanding framework and our results highlight a direct pathway to handling long contexts in LLMs with carefully managed chunking and aggregator strategies.

Jingfeng Chen, Raghuveer Thirukovalluru, Junlin Wang et al.

Large Language Models (LLMs) often produce factoid hallucinations - plausible yet incorrect answers. A common mitigation strategy is model alignment, which improves factual accuracy by training on curated (factual, non-factual) pairs. However, this approach often relies on a stronger model (e.g., GPT-4) or an external knowledge base to assess factual correctness that may not always be accessible. Addressing this, we propose Atomic Consistency Preference Optimization (ACPO), a self-supervised preference-tuning method that enhances factual accuracy without external supervision. ACPO leverages atomic consistency signals (i.e., the agreement of individual facts across multiple stochastic responses) to identify high- and low-quality data pairs for model alignment. Despite being fully self-supervised, ACPO outperforms the strong supervised alignment baseline by 1.95 points averaged across Phi-3 and Llama3 on the LongFact and BioGen datasets, demonstrating its effectiveness in improving factual reliability without relying on external models or knowledge bases.

Roy Xie, Chengxuan Huang, Junlin Wang et al.

Large language models (LLMs) have significantly transformed the educational landscape. As current plagiarism detection tools struggle to keep pace with LLMs' rapid advancements, the educational community faces the challenge of assessing students' true problem-solving abilities in the presence of LLMs. In this work, we explore a new paradigm for ensuring fair evaluation -- generating adversarial examples which preserve the structure and difficulty of the original questions aimed for assessment, but are unsolvable by LLMs. Focusing on the domain of math word problems, we leverage abstract syntax trees to structurally generate adversarial examples that cause LLMs to produce incorrect answers by simply editing the numeric values in the problems. We conduct experiments on various open- and closed-source LLMs, quantitatively and qualitatively demonstrating that our method significantly degrades their math problem-solving ability. We identify shared vulnerabilities among LLMs and propose a cost-effective approach to attack high-cost models. Additionally, we conduct automatic analysis to investigate the cause of failure, providing further insights into the limitations of LLMs.

Jiancheng Fang, Shaoyu Wang, Junlin Wang et al.

Multi-source stationary computed tomography (CT) has recently attracted attention for its ability to achieve rapid image reconstruction, making it suitable for time-sensitive clinical and industrial applications. However, practical systems are often constrained by ultra-sparse-view sampling, which significantly degrades reconstruction quality. Traditional methods struggle under ultra-sparse-view settings, where interpolation becomes inaccurate and the resulting reconstructions are unsatisfactory. To address this challenge, this study proposes Diffusion-Refined Neural Attenuation Fields (Diff-NAF), an iterative framework tailored for multi-source stationary CT under ultra-sparse-view conditions. Diff-NAF combines a Neural Attenuation Field representation with a dual-branch conditional diffusion model. The process begins by training an initial NAF using ultra-sparse-view projections. New projections are then generated through an Angle-Prior Guided Projection Synthesis strategy that exploits inter view priors, and are subsequently refined by a Diffusion-driven Reuse Projection Refinement Module. The refined projections are incorporated as pseudo-labels into the training set for the next iteration. Through iterative refinement, Diff-NAF progressively enhances projection completeness and reconstruction fidelity under ultra-sparse-view conditions, ultimately yielding high-quality CT reconstructions. Experimental results on multiple simulated 3D CT volumes and real projection data demonstrate that Diff-NAF achieves the best performance under ultra-sparse-view conditions.

Junlin Wang, Jue Wang, Zhen et al.

Recent advances in large language models (LLMs) opened up new directions for leveraging the collective expertise of multiple LLMs. These methods, such as Mixture-of-Agents, typically employ additional inference steps to generate intermediate outputs, which are then used to produce the final response. While multi-agent inference can enhance response quality, it can significantly increase the time to first token (TTFT), posing a challenge for latency-sensitive applications and hurting user experience. To address this issue, we propose staircase streaming for low-latency multi-agent inference. Instead of waiting for the complete intermediate outputs from previous steps, we begin generating the final response as soon as we receive partial outputs from these steps. Experimental results demonstrate that staircase streaming reduces TTFT by up to 93% while maintaining response quality.

Junlin Wang, Zehao Wu, Shaowei Lu et al.

Retrieval-Augmented Generation (RAG) enables large language models (LLMs) to access external knowledge sources, but the effectiveness of RAG relies on the coordination between the retriever and the generator. Since these components are developed independently, their interaction is often suboptimal: the retriever may return irrelevant or redundant documents, while the generator may fail to fully leverage retrieved evidence. In this work, we propose a process-supervised multi-agent framework to bridge the gap between retriever and generator. The framework introduces two lightweight agents: a Decision Maker, which determines when to continue retrieval or stop for answer generation, and a Knowledge Selector, which filters retrieved documents to retain only the most useful evidence. To provide fine-grained supervision, we employ an LLM-as-a-Judge that evaluates each intermediate action with process-level rewards, ensuring more accurate credit assignment than relying solely on final answer correctness. We further adopt a tree-structured rollout strategy to explore diverse reasoning paths, and train both agents with Proximal Policy Optimization (PPO) in an end-to-end manner. Experiments on single-hop and multi-hop question answering benchmarks show that our approach achieves higher accuracy, more stable convergence, and produces more interpretable reasoning trajectories compared with standard RAG baselines. Importantly, the proposed framework is modular and plug-and-play, requiring no modification to the retriever or generator, making it practical for real-world RAG applications.

Zonghan Wu, Congyuan Zou, Junlin Wang et al.

Generative AI, particularly large language models (LLMs), is beginning to transform the financial industry by automating tasks and helping to make sense of complex financial information. One especially promising use case is the automatic creation of fundamental analysis reports, which are essential for making informed investment decisions, evaluating credit risks, guiding corporate mergers, etc. While LLMs attempt to generate these reports from a single prompt, the risks of inaccuracy are significant. Poor analysis can lead to misguided investments, regulatory issues, and loss of trust. Existing financial benchmarks mainly evaluate how well LLMs answer financial questions but do not reflect performance in real-world tasks like generating financial analysis reports. In this paper, we propose FinAR-Bench, a solid benchmark dataset focusing on financial statement analysis, a core competence of fundamental analysis. To make the evaluation more precise and reliable, we break this task into three measurable steps: extracting key information, calculating financial indicators, and applying logical reasoning. This structured approach allows us to objectively assess how well LLMs perform each step of the process. Our findings offer a clear understanding of LLMs current strengths and limitations in fundamental analysis and provide a more practical way to benchmark their performance in real-world financial settings.

Roy Xie, Junlin Wang, Paul Rosu et al.

Large language models (LLMs) process entire input contexts indiscriminately, which is inefficient when the information required to answer a query is localized within the context. We present dynamic context cutoff, a novel method enabling LLMs to self-terminate processing upon acquiring sufficient task-relevant information. Through analysis of model internals, we discover that specific attention heads inherently encode "sufficiency signals" -- detectable through lightweight classifiers -- that predict when critical information has been processed. This reveals a new efficiency paradigm: models' internal understanding naturally dictates processing needs rather than external compression heuristics. Comprehensive experiments across six QA datasets (up to 40K tokens) with three model families (LLaMA/Qwen/Mistral, 1B-70B) demonstrate 3.4% accuracy improvement while achieving 1.33x token reduction on average. Furthermore, our method demonstrates superior performance compared to other context efficiency methods at equivalent token reduction rates. Additionally, we observe an emergent scaling phenomenon: while smaller models require probing for sufficiency detection, larger models exhibit intrinsic self-assessment capabilities through prompting.

Roy Xie, Junlin Wang, Ruomin Huang et al.

The rapid scaling of large language models (LLMs) has raised concerns about the transparency and fair use of the data used in their pretraining. Detecting such content is challenging due to the scale of the data and limited exposure of each instance during training. We propose ReCaLL (Relative Conditional Log-Likelihood), a novel membership inference attack (MIA) to detect LLMs' pretraining data by leveraging their conditional language modeling capabilities. ReCaLL examines the relative change in conditional log-likelihoods when prefixing target data points with non-member context. Our empirical findings show that conditioning member data on non-member prefixes induces a larger decrease in log-likelihood compared to non-member data. We conduct comprehensive experiments and show that ReCaLL achieves state-of-the-art performance on the WikiMIA dataset, even with random and synthetic prefixes, and can be further improved using an ensemble approach. Moreover, we conduct an in-depth analysis of LLMs' behavior with different membership contexts, providing insights into how LLMs leverage membership information for effective inference at both the sequence and token level.

Junlin Wang, Siddhartha Jain, Dejiao Zhang et al.

A diverse array of reasoning strategies has been proposed to elicit the capabilities of large language models. However, in this paper, we point out that traditional evaluations which focus solely on performance metrics miss a key factor: the increased effectiveness due to additional compute. By overlooking this aspect, a skewed view of strategy efficiency is often presented. This paper introduces a framework that incorporates the compute budget into the evaluation, providing a more informative comparison that takes into account both performance metrics and computational cost. In this budget-aware perspective, we find that complex reasoning strategies often don't surpass simpler baselines purely due to algorithmic ingenuity, but rather due to the larger computational resources allocated. When we provide a simple baseline like chain-of-thought self-consistency with comparable compute resources, it frequently outperforms reasoning strategies proposed in the literature. In this scale-aware perspective, we find that unlike self-consistency, certain strategies such as multi-agent debate or Reflexion can become worse if more compute budget is utilized.

Phillip Howard, Junlin Wang, Vasudev Lal et al.

Comparative knowledge (e.g., steel is stronger and heavier than styrofoam) is an essential component of our world knowledge, yet understudied in prior literature. In this paper, we harvest the dramatic improvements in knowledge capabilities of language models into a large-scale comparative knowledge base. While the ease of acquisition of such comparative knowledge is much higher from extreme-scale models like GPT-4, compared to their considerably smaller and weaker counterparts such as GPT-2, not even the most powerful models are exempt from making errors. We thus ask: to what extent are models at different scales able to generate valid and diverse comparative knowledge? We introduce NeuroComparatives, a novel framework for comparative knowledge distillation overgenerated from language models such as GPT-variants and LLaMA, followed by stringent filtering of the generated knowledge. Our framework acquires comparative knowledge between everyday objects, producing a corpus of up to 8.8M comparisons over 1.74M entity pairs - 10X larger and 30% more diverse than existing resources. Moreover, human evaluations show that NeuroComparatives outperform existing resources in terms of validity (up to 32% absolute improvement). Our acquired NeuroComparatives leads to performance improvements on five downstream tasks. We find that neuro-symbolic manipulation of smaller models offers complementary benefits to the currently dominant practice of prompting extreme-scale language models for knowledge distillation.

Junchen Zhao, Yurun Song, Junlin Wang et al.

Automatic code generation from natural language descriptions can be highly beneficial during the process of software development. In this work, we propose GAP-Gen, a Guided Automatic Python Code Generation method based on Python syntactic constraints and semantic constraints. We first introduce Python syntactic constraints in the form of Syntax-Flow, which is a simplified version of Abstract Syntax Tree (AST) reducing the size and high complexity of Abstract Syntax Tree but maintaining crucial syntactic information of Python code. In addition to Syntax-Flow, we introduce Variable-Flow which abstracts variable and function names consistently through out the code. In our work, rather than pretraining, we focus on modifying the finetuning process which reduces computational requirements but retains high generation performance on automatic Python code generation task. GAP-Gen fine-tunes the transformer based language models T5 and CodeT5 using the Code-to-Docstring datasets CodeSearchNet, CodeSearchNet AdvTest and Code-Docstring Corpus from EdinburghNLP. Our experiments show that GAP-Gen achieves better results on automatic Python code generation task than previous works.

Junlin Wang, Jens Tuyls, Eric Wallace et al.

Gradient-based analysis methods, such as saliency map visualizations and adversarial input perturbations, have found widespread use in interpreting neural NLP models due to their simplicity, flexibility, and most importantly, their faithfulness. In this paper, however, we demonstrate that the gradients of a model are easily manipulable, and thus bring into question the reliability of gradient-based analyses. In particular, we merge the layers of a target model with a Facade that overwhelms the gradients without affecting the predictions. This Facade can be trained to have gradients that are misleading and irrelevant to the task, such as focusing only on the stop words in the input. On a variety of NLP tasks (text classification, NLI, and QA), we show that our method can manipulate numerous gradient-based analysis techniques: saliency maps, input reduction, and adversarial perturbations all identify unimportant or targeted tokens as being highly important. The code and a tutorial of this paper is available at http://ucinlp.github.io/facade.

Eric Wallace, Jens Tuyls, Junlin Wang et al.

Neural NLP models are increasingly accurate but are imperfect and opaque---they break in counterintuitive ways and leave end users puzzled at their behavior. Model interpretation methods ameliorate this opacity by providing explanations for specific model predictions. Unfortunately, existing interpretation codebases make it difficult to apply these methods to new models and tasks, which hinders adoption for practitioners and burdens interpretability researchers. We introduce AllenNLP Interpret, a flexible framework for interpreting NLP models. The toolkit provides interpretation primitives (e.g., input gradients) for any AllenNLP model and task, a suite of built-in interpretation methods, and a library of front-end visualization components. We demonstrate the toolkit's flexibility and utility by implementing live demos for five interpretation methods (e.g., saliency maps and adversarial attacks) on a variety of models and tasks (e.g., masked language modeling using BERT and reading comprehension using BiDAF). These demos, alongside our code and tutorials, are available at https://allennlp.org/interpret .

Junlin Wang, Xiao Li, Tzu-Hao Huang et al.

This paper presents design principles for comfort-centered wearable robots and their application in a lightweight and backdrivable knee exoskeleton. The mitigation of discomfort is treated as mechanical design and control issues and three solutions are proposed in this paper: 1) a new wearable structure optimizes the strap attachment configuration and suit layout to ameliorate excessive shear forces of conventional wearable structure design; 2) rolling knee joint and double-hinge mechanisms reduce the misalignment in the sagittal and frontal plane, without increasing the mechanical complexity and inertia, respectively; 3) a low impedance mechanical transmission reduces the reflected inertia and damping of the actuator to human, thus the exoskeleton is highly-backdrivable. Kinematic simulations demonstrate that misalignment between the robot joint and knee joint can be reduced by 74% at maximum knee flexion. In experiments, the exoskeleton in the unpowered mode exhibits 1.03 Nm root mean square (RMS) low resistive torque. The torque control experiments demonstrate 0.31 Nm RMS torque tracking error in three human subjects.