论文检索

Deep Learning · Large Language Models

Howard Chen, Noam Razin, Karthik Narasimhan, Danqi Chen

Adapting language models (LMs) to new tasks via post-training carries the risk of degrading existing capabilities -- a phenomenon classically known as catastrophic forgetting. In this paper, toward identifying guidelines for mitigating this phenomenon, we systematically compare the forgetting patterns of two widely adopted post-training methods: supervised fine-tuning (SFT) and reinforcement learning (RL). Our experiments reveal a consistent trend across LM families (Llama, Qwen) and tasks (instruction following, general knowledge, and arithmetic reasoning): RL leads to less forgetting than SFT while achieving comparable or higher target task performance. To investigate the cause for this difference, we consider a simplified setting in which the LM is modeled as a mixture of two distributions, one corresponding to prior knowledge and the other to the target task. We identify that the mode-seeking nature of RL, which stems from its use of on-policy data, enables keeping prior knowledge intact when learning the target task. We then verify this insight by demonstrating that the use on-policy data underlies the robustness of RL to forgetting in practical settings, as opposed to other algorithmic choices such as the KL regularization or advantage estimation. Lastly, as a practical implication, our results highlight the potential of mitigating forgetting using approximately on-policy data, which can be substantially more efficient to obtain than fully on-policy data.

Social Aspects · Everything Else

Zijin Chen, lesui Yu, Xiaofei Liao, Hai Jin, Qinbin Li

As submission volumes grow, AI conference peer review increasingly suffers from scale drift and non-comparable scoring: similar rationales can yield markedly different numeric ratings due to subjective calibration and occasional incoherent or strategic scoring, even though scores often strongly influence outcomes. This position paper argues that **AI conference workflows should incorporate an LLM-driven calibration layer that maps reviewer rationales (e.g., strengths and weaknesses) into consistent and auditable anchor scores**. The residual between a reviewer’s reported score and the anchor score turns rationale--score misalignment into a measurable signal for targeted escalation. We instantiate an end-to-end pipeline and apply it to OpenReview data from ICLR 2023--2025 to quantify severity/leniency patterns and where misalignment concentrates. We further propose a lightweight post-check---requesting added justification or score revision when residuals are large---and estimate its impact via an offline counterfactual simulation. Finally, we outline an adoption playbook and governance boundaries, emphasizing that the LLM audits scoring coherence rather than replacing human judgment or making accept/reject decisions.

General Machine Learning · Evaluation

Zachary Berger, Daniel Prakah-Asante, John Guttag, Collin Stultz

This position paper argues that current benchmarking practice in 12-lead ECG representation learning must be fixed to ensure progress is reliable and aligned with clinically meaningful objectives. The field has largely converged on three public multi-label benchmarks (PTB-XL, CPSC2018, CSN) dominated by arrhythmia and waveform-morphology labels, even though the ECG is known to encode substantially broader clinical information. We argue that downstream evaluation should expand to include an assessment of structural heart disease and patient-level forecasting, in addition to other evolving ECG-related endpoints, as relevant clinical targets. Next, we outline evaluation best practices for multi-label, imbalanced settings, and show that when they are applied, the literature's current conclusion about which representations perform best is altered. Furthermore, we demonstrate the surprising result that a randomly initialized encoder with linear evaluation matches state-of-the-art pre-training on many tasks. This motivates the use of a random encoder as a reasonable baseline model. We substantiate our observations with an empirical evaluation of three representative ECG pre-training approaches across six evaluation settings: the three standard benchmarks, a structural disease dataset, hemodynamic inference, and patient forecasting.

Deep Learning · Large Language Models

Shiqiang Wang, Herbert Woisetschlaeger, Hans-Arno Jacobsen, Mingyue Ji

Data is fundamental to large language models (LLMs). However, understanding of what makes certain data useful for different stages of an LLM workflow, including training, tuning, alignment, in-context learning, etc., and why, remains an open question. Current approaches rely heavily on extensive experimentation with large public datasets to obtain empirical heuristics for data filtering and dataset construction. These approaches are compute intensive and lack a principled way of understanding the essence of how specific data characteristics drive LLM behavior. In this position paper, *we advocate for the need of developing systematic methodologies for generating synthetic sequences from appropriately defined random processes, with the goal that these sequences can reveal useful characteristics when they are used in one or multiple stages of the LLM workflow*. We refer to such sequences as *data probes*. By observing LLM behavior on data probes, researchers can systematically conduct studies on how data characteristics influence model performance, generalization, and robustness. The probing sequences exhibit statistical properties that can be viewed using theoretical concepts, such as typical sets, which are generalized to describe the behaviors of LLMs. This data-probe approach provides a pathway for uncovering foundational insights into the role of data in LLM training and inference, beyond empirical heuristics.

General Machine Learning · Causality

Ruta Binkyte, Ivaxi Sheth, Zhijing Jin, Mohammad Havaei, Bernhard Schölkopf, Mario Fritz

As artificial intelligence (AI), including machine learning (ML) models and foundation models (FMs), is increasingly deployed in high-stakes domains, ensuring their trustworthiness has become a central challenge. However, the core trustworthy AI objectives, such as fairness, robustness, privacy, and explainability, are hard to achieve simultaneously, especially while preserving utility. This position paper argues that causality is necessary to understand and balance trade-offs in performance and multiple objectives of trustworthy AI. We ground our arguments in re-interpreting trustworthy AI trade-offs as incompatible invariance requirements under different changes to the data-generating process. We then illustrate that causality provides a unifying framework for understanding how trade-offs in trustworthy AI arise, and how they can be softened or resolved through selective invariance. This perspective applies to both classical ML models and large-scale FMs. Our paper discusses how causal assumptions may be applied explicitly or implicitly in modern large-scale systems. Finally, we outline open challenges and opportunities for using causality to build more trustworthy AI.

Deep Learning · Large Language Models

Zongqi Wang, Rui Wang, Yuchuan Wu, Yiyao Yu, Pinyi Zhang, Shaoning Sun, Yujiu Yang, Yongbin Li

Reinforcement Learning with Verifiable Reward (RLVR) on preference data has become the mainstream approach for training Generative Reward Models (GRMs). Typically, GRMs generate reasoning chains ending with critiques and preference labels, with RLVR using label correctness as the training reward. However, we demonstrate that such binary classification tasks make GRMs susceptible to guessing correct outcomes without sound critiques, introducing noise into the reward signal and impairing learning effectiveness. To address this, we propose Reward Modeling from Natural Language Human Feedback (RM-NLHF), which leverages natural language feedback to obtain process reward signals. Specifically, we compute the similarity between GRM-generated and human critiques as the process reward, providing more accurate signals than outcome-only supervision. Considering that human critiques are difficult to scale, we introduce MetaRM which learns to predict process reward from datasets with human critiques and generalizes to data without them. Experiments on multiple benchmarks demonstrate that RM-NLHF consistently outperforms state-of-the-art models trained with outcome reward, confirming the superiority of natural language over binary feedback.

Social Aspects · Safety

Janis Keuper, Margret Keuper

The wide availability and low usability barrier of modern image generation models has triggered the reasonable fear of criminal misconduct and negative social implications. The machine learning community has been engaging this problem with an extensive series of publications proposing algorithmic solutions for the detection of "fake'', e.g. entirely generated or partially manipulated images. While there is undoubtedly some progress towards technical solutions of the problem, we argue that current and prior work is focusing too much on generative algorithms and "fake'' data-samples, neglecting a clear definition and data collection of "real'' images. The fundamental question *"what is a real image?''* might appear to be quite philosophical, but our analysis shows that the development and evaluation of basically all current "fake''-detection methods is relying on only a few, quite old low-resolution datasets of "real'' images like *ImageNet*. However, the technology for the acquisition of "real'' images, aka taking photos, has drastically evolved over the last decade: Today, over 90% of all photographs are produced by smartphones which typically use algorithms to compute an image from multiple inputs (over time) from multiple sensors. Based on the fact that these image formation algorithms are typically neural network architectures which are closely related to "fake''-image generators, we state the position that today, **we need to re-think the concept of "real'' images**. The purpose of this position paper is to raise the awareness of the current shortcomings in this active field of research and to trigger an open discussion whether the detection of ``fake'' images is a sound objective at all. At the very least, we need a clear technical definition of "real'' images and new benchmark datasets.

Sobhan Lotfi, Ava Iranmanesh, Lachin Naghashyar, Ali Shirali, Fateme Haredasht, Sanmi Koyejo, Phil Torr, Yong Suk Lee, Fazl Barez, Joel Lehman 等

Breakthroughs often come from ideas we could not have predicted in advance. In biology, this is called exaptation: traits evolved for one function become decisive for another. Scientific progress works similarly, but only if ideas survive periods when they appear uncompetitive by current metrics. This position paper argues that AI's benchmark-centered selection environment, while successful at bypassing complex debates about the nature of intelligence, taxes exaptation. When one selection rule dominates, ideas that do not fit it have nowhere to persist. The cost grows acute as the field shifts from asking can machines exhibit intelligent behavior? to asking can machines exhibit intelligent behavior such that they are aligned, interpretable, and safe? These are philosophically distinct questions that may require discoveries that we cannot specify. We propose mechanisms to restore exaptive capacity without abandoning benchmarking: plural evaluation regimes, protected venues for non-comparable work, long-horizon funding, and training norms that encourage researchers to question selection rules, not only optimize within them.

Social Aspects · Accountability, Transparency, and Interpretability

Azmine Toushik Wasi, Mst Islam, Mahfuz Anik, Manjurul Ahsan, Taki Hasan Rafi, Dong-Kyu Chae

As Artificial Intelligence (AI) becomes increasingly embedded in global infrastructure, the urgency for robust governance frameworks has intensified. However, current approaches, led by jurisdiction-specific laws such as the EU AI Act, China's algorithm governance, and the NIST AI Risk Management Framework in the U.S., create a fragmented regulatory landscape. In this position paper, we argue that \textbf{\textit{AI governance must be built not on laws alone, but on ISO-like interoperability protocols that enable standardized, machine-readable risk communication across borders}}. Drawing on the success of the GDPR, which was operationalized through standards like ISO 27001 and Privacy by Design, we propose the development of standardized AI \textit{nutrition labels} containing unified metrics for bias, energy usage, and data provenance to facilitate cross-jurisdictional compliance. These manifests would lower barriers for small and medium enterprises (SMEs), reduce redundant regulatory efforts, and build public trust. The paper addresses concerns that standards may stifle innovation by advocating for modular, versioned protocols designed to evolve in tandem with technological change. Overall, we call for a shift from siloed legal compliance toward interoperable technical conformance, enabling a shared global language for responsible AI deployment.

Deep Learning · Large Language Models

Yingxuan Zhuang, Miao Pan, Jingxiao Yang, Jintao Chen, Cheng Tan, Yuxiang Cai, Siwei Tan, Chen Zhi, Xuhong Zhang, Jianwei Yin

Multimodal Large Language Models often suffer from object hallucinations, where generated outputs are inconsistent with the visual evidence. This issue is typically attributed to the over-reliance on language priors, which can override the visual context. Recent training-free decoding strategies address this by penalizing language priors. However, these methods overlook the dual nature of language priors, where they can be both helpful and harmful depending on the alignment with visual evidence. In particular, blindly suppressing language priors often disrupts the model’s semantic manifold, leading to performance degradation, a phenomenon we term Manifold Departure. To address this, we propose Manifold-Guided Adaptive Projection (MGAP), a geometry-aware, training-free decoding method that mitigates hallucinations while preserving representation structure. MGAP first constructs a language-prior subspace from blind hidden states (null-image inputs) via SVD. During decoding, MGAP projects each multimodal hidden state onto this subspace and applies a consistency-aware gate to adaptively attenuate only the projected prior component, yielding a subspace-selective update that largely preserves the orthogonal semantic components. Extensive experiments on POPE and CHAIR show that MGAP outperforms prior decoding baselines, achieving stronger hallucination suppression without sacrificing coherence.

Social Aspects · Accountability, Transparency, and Interpretability

Matthew Riemer, Tommaso Tosato, Maximilian Puelma Touzel, Amin Memarian, Guillaume Dumas, Glen Berseth, Irina Rish

This position paper argues that AI agents with chain-of-thought reasoning capabilities are predisposed to exhibit collusive behavior and should be required to obtain behavioral certification before making decisions that affect economic markets. This is because integrating these agents into society could collapse the legal evidentiary distinction between competition and collusion among independent firms without eroding the economic harm distinction. Experiments with DeepSeek-R1 agents in the Bertrand oligopoly pricing domain reveal a tendency towards tacit collusion that persists even when humans prompt the agents not to collude. We further show that the chain-of- thought of these agents can be steered toward either extremely collusive or highly competitive behavior in a way that is not semantically detectable by another LLM analyzing the reasoning traces. As a result, deploying reasoning agents for market decisions leads to collusive economic outcomes without any evidence of conspiracy or intent. Thus, certification based on observed behavior in representative situations is necessary to prevent collusion. We provide preliminary evidence that such agents can be steered in a generalizable way toward efficient competitive equilibria. However, developing a comprehensive behavioral certification will be required before these models can be deployed in real-world markets while ensuring their stability and efficiency.

Applications · Computer Vision

Nedyalko Prisadnikov, Danda Pani Paudel, Yuqian Fu, Luc Van Gool

This position paper argues that the next generation of vision encoders should be image size agnostic and task driven. The source of our inspiration is biological. Not a structural aspect of biological vision, but a behavioral trait – efficiency. We focus on a couple of ways in which vision in nature is efficient, but modern vision encoders not. We – humans and animals – deal with vast quantities of visual data, and need to be smart where we focus our limited energy – it depends on the task. It is our belief that vision encoders should be dynamic and the computational complexity should depend on the task at hand rather than the size of the image. We, also, provide concrete first steps towards our vision – a proof-of-concept solution for image classification. Despite classification being not very representative for what we are trying to achieve, it shows that our approach is feasible and promising.

Deep Learning · Large Language Models

Yang Ouyang, Shuhang Lin, Jung-Eun Kim

Large language models (LLMs) demonstrate strong chain-of-thought (CoT) reasoning abilities, while smaller models ($\leq$ 3B parameters) significantly underperform on multi-step reasoning tasks. Based on empirical analyses of the Qwen-2.5 model family on math reasoning benchmarks, we find that more proficient reasoning is associated with fewer reasoning steps but higher information density per step, a property we term *Dense Reasoning*. Motivated by this observation, we propose **DenseSteer**, a training-free inference-time steering framework that enhances small-model reasoning by modulating internal representations toward dense reasoning patterns. Experiments show that our method yields consistent accuracy improvements without increasing token-level Negative Log-Likelihood, highlighting dense reasoning as an effective structural approach to mathematical problem solving.

Deep Learning · Large Language Models

Yulin Chen, He He, Chen Zhao

Reinforcement Learning with Verifiable Reward (RLVR) has proven effective in improving Large Language Model's (LLM) reasoning ability. However, the learning dynamics of RLVR remain underexplored. In this paper, we reveal a curious phenomenon: among hard examples that the model initially struggles with, a substantial subset remains unlearnable even when correct rollouts present. To understand the phenomenon, we first demonstrate that existing optimization and sampling techniques fail to resolve unlearnability. With cross-example gradient analysis, we show that unlearnable examples have fundamental representation issue, characterized by low gradient similarity with the rest of the examples and ungeneralizable reasoning patterns. We further show that representation flaws are difficult to mitigate in RL, as data augmentation does not improve gradient similarity. Our study provides the first systematic characterization of unlearnable data in RLVR training and reveals fundamental limitations in current RL approaches for reasoning tasks.

Applications · Time Series

Md Asif Bin Syed, Md Younus Ahamed, Azmine Toushik Wasi

Time series foundation models (TSFMs) have demonstrated strong performance on established benchmarks such as GIFT-Eval, Monash, and TSFM-Bench. However, these benchmarks pool datasets from many domains with uneven representation, which can obscure performance within specific application areas such as healthcare, finance, nature, retail, and transport. The necessity for domain-specific evaluation arises from the inherent structural diversity of time series data: clinical records often feature irregular sampling and informative missingness; financial sequences are characterized by high noise and stochastic trajectories; and environmental data, such as energy and weather, are governed by deterministic physical laws and strong seasonal hierarchies. Motivated by this heterogeneity, **we argue that TSFMs require explicit domain-specific benchmarks** so practitioners can reliably assess a model's utility within their own application area. This is because cross-domain differences in data generation, sampling irregularity, and nonstationarity under concept drift fundamentally shape forecasting difficulty and failure modes. As a result, strong performance on aggregated leaderboards may not translate to reliable deployment within a specific domain. To test this, we evaluated seven TSFMs across 72 datasets from six domains (healthcare, finance, energy, nature, transport, and retail) and found substantial cross-domain variability. These findings confirm that global benchmark scores can be misleading and that domain-aware evaluations are essential for trustworthy TSFM selection.

Deep Learning · Large Language Models

Shubham Parashar, Atharv Chagi, Jacob Helwig, Lakshmi Madhavarapu, Sushil Vemuri, James Caverlee, Dileep Kalathil, Shuiwang Ji

We aim to improve the reasoning capabilities of diffusion language models (DLMs). While SFT performs well for autoregressive models, its use in DLMs faces challenges. Our observation and analysis reveal that vanilla SFT does not consider learnability, i.e., what and when tokens are learned. Specifically, we observe that rare tokens are difficult to learn when most of the input is masked. In contrast, it is straightforward and thus of little value to learn common tokens when most of the input is unmasked. To consider learnability, we propose LIFT, a learnability-informed fine-tuning strategy for DLMs. LIFT learns easy tokens when most of the input is noisy and hard tokens when more input is available, thereby aligning training with information available at different diffusion time steps. Our results show that LIFT outperforms existing SFT baselines across six reasoning benchmarks, achieving up to a 3x relative gain on AIME'24 and AIME'25.

Deep Learning · Generative Models and Autoencoders

Onkar Susladkar, Tushar Prakash, Gayatri Deshmukh, Kiet Nguyen, Jiaxun Zhang, Adheesh Juvekar, Tianshu Bao, Lin Chai, Sparsh Mittal, Inderjit Dhillon 等

We propose UniDFlow, a unified discrete flow-matching framework for multimodal understanding, generation, and editing. It decouples understanding and generation via task-specific low-rank adapters, avoiding objective interference and representation entanglement, while a novel reference-based multimodal preference alignment optimizes relative outcomes under identical conditioning, improving faithfulness and controllability without large-scale retraining. UniDFlow achieves SOTA performance across eight benchmarks and exhibits strong zero-shot generalization to tasks including inpainting, in-context image generation, reference-based editing, and compositional generation, despite no explicit task-specific training.

Reinforcement Learning · Everything Else

Boyang Xu, Qing Zou, Siqin Yang, Hao Yan

Distributional RL models the full return distribution, but common categorical/quantile approaches rely on projection and independently sampled Bellman targets, which ignore the Bellman operator’s affine transport structure and yield high-variance learning signals. We introduce Path-Coupled Bellman Flows, a flow-matching framework that shares base noise to couple the generative trajectories of consecutive states, inducing a geometric Bellman scaling law between their velocity fields. This geometry motivates a $\lambda$-family of Bellman-flow objectives that functions as a control variate, reducing variance while retaining the same Bellman-consistent fixed point. Across toy diagnostics and offline RL benchmarks (OGBench, D4RL), our method improves training stability and achieves competitive or improved performance relative to prior distributional baselines.

Deep Learning · Algorithms

Yechan Kang, Yongjin Kweon, Mingyeong Seo, Sohee Park, Yeonguk jeon, Jongkil Park, Hyun Jae Jang, Jaewook Kim, Yeonjoo Jeong, Suyoun Lee 等

Training deep spiking neural networks (SNNs) remains challenging due to sharp loss landscapes and temporal inconsistency caused by surrogate gradients. To address these challenges, we propose a unified framework: adaptive and asymmetric surrogate gradients (A$^2$SG). The adaptive gradients adjust an effective window for spatio-temporal adaptation, reducing spatial gradient variation and maintaining directional consistency of gradients over time. The asymmetric gradients reflect neuronal dynamics by assigning larger gradients to neurons with higher membrane potentials, and we prove that they yield lower variation than symmetric surrogates. Our analysis further establishes a direct connection between local gradient variation and the curvature of the loss landscape, providing a principled explanation for how A$^2$SG promotes convergence to flatter minima and improves generalization. We conduct extensive experiments on diverse models, including CNN-based and Transformer-based SNNs, across various tasks such as image classification using both static and neuromorphic datasets, as well as segmentation. The results demonstrate that A$^2$SG consistently improves accuracy and energy efficiency, establishing it as a general and reliable solution for training deep SNNs.

Deep Learning · Large Language Models

Elron Bandel, Asaf Yehudai, Alexandre Lacoste, Avijit Ghosh, Graham Neubig, Margaret Mitchell, Michal Shmueli-Scheuer, Leshem Choshen

We call for the development of agentic systems that thrive in new environments. Agentic systems, comprising foundation models, tools, and an execution strategy, have demonstrated strong capabilities, yet their development is often constrained by narrow benchmarks and their operation is siloed to limited environments. This paper advocates for developing general, adaptive agents that excel across diverse environments, from terminals and web interfaces to biological and embodied settings. We examine current limitations, explain the potential of increased generality, and identify immediate development priorities. Finally, we argue that protocols and evaluation must prioritize adaptiveness to foster a shared ecosystem for general-purpose agentic systems.