论文检索

Social Aspects · Trustworthy Machine Learning

Justin Lee, Zheda Mai, Jinsu Yoo, Chongyu Fan, Cheng Zhang, Wei-Lun Chao

Machine unlearning—the ability to remove designated concepts from a pre-trained model—has advanced rapidly, particularly for text-to-image diffusion models. However, existing methods typically assume that unlearning requests arrive all at once, whereas in practice they often arrive sequentially. We present the first systematic study of continual unlearning in text-to-image diffusion models and show that popular unlearning methods suffer from rapid utility collapse: after only a few requests, models forget retained knowledge and generate degraded images. We trace this failure to cumulative parameter drift from the pre-training weights and argue that regularization is crucial to addressing it. To this end, we study a suite of add-on regularizers that (1) mitigate drift and (2) remain compatible with existing unlearning methods. Beyond generic regularizers, we show that semantic awareness is essential for preserving concepts close to the unlearning target, and propose a gradient-projection method that constrains parameter drift orthogonal to their subspace. This substantially improves continual unlearning performance and is complementary to other regularizers for further gains. Taken together, our study establishes continual unlearning as a fundamental challenge in text-to-image generation and provides insights, baselines, and open directions for advancing safe and accountable generative AI.

Social Aspects · Fairness, Equity, Justice and Safety

Alexander Panfilov, Evgenii Kortukov, Kristina Nikolić, Matthias Bethge, Sebastian Lapuschkin, Wojciech Samek, Ameya Prabhu, Maksym Andriushchenko, Jonas Geiping

Large language model (LLM) developers aim for their models to be honest, helpful, and harmless. However, when faced with malicious requests, models are trained to refuse, sacrificing helpfulness. We show that frontier LLMs can develop a preference for \textit{dishonesty} as a new strategy, even when other options are available. Affected models respond to harmful requests with outputs that sound harmful but are crafted to be subtly incorrect or otherwise harmless in practice. This behavior emerges with hard-to-predict variations even within models from the same model family. We find no apparent cause for the propensity to deceive, but show that more capable models are better at executing this strategy. Strategic dishonesty already has a practical impact on safety evaluations, as we show that dishonest responses fool \emph{all} output-based monitors used to detect jailbreaks that we test, rendering benchmark scores unreliable. Further, strategic dishonesty can act like a \emph{honeypot} against malicious users, which noticeably obfuscates prior jailbreak attacks. While output monitors fail, we show that linear probes on internal activations can be used to reliably detect strategic dishonesty. We validate probes on datasets with verifiable outcomes and by using them as steering vectors. Overall, we consider strategic dishonesty as a concrete example of a broader concern that alignment of LLMs is hard to control, especially when helpfulness and harmlessness conflict.

Yeongmin Kim, Donghyeok Shin, Mina Kang, Byeonghu Na, Il-chul Moon

Large language models (LLMs) deliver remarkable performance but are costly to deploy, motivating knowledge distillation (KD) for efficient inference. Existing KD objectives typically match student and teacher probabilities via softmax, which blurs valuable logit information. While direct logit distillation (DLD) mitigates softmax smoothing, it fails to account for logit shift invariance, thereby restricting the solution space. We propose Concrete Score Distillation (CSD), a discrete score-matching objective that overcomes both softmax-induced smoothing and restrictions on the optimal solution set. We resolve the training instability and quadratic complexity of discrete score-matching in autoregressive LLMs, and the resulting CSD objective aligns relative logit differences across all vocabulary pairs between student and teacher with flexible weighting. We provide both mode-seeking and mode-covering instances within our framework and evaluate CSD on task-agnostic instruction-following, task-specific, and general chat capability distillation using GPT-2-1.5B, OpenLLaMA-7B, and Gemma-7B-IT, Qwen2.5-7B-IT, and Gemma2-9B-IT teachers. Experiments show that CSD consistently surpasses recent KD objectives, achieves favorable fidelity–diversity trade-offs, and yields complementary gains when combined with on-policy techniques, demonstrating its scalability and effectiveness for LLM distillation. Code: https://github.com/aailab-kaist/CSD.

Social Aspects · Fairness, Equity, Justice and Safety

Mikhail Terekhov, Alexander Panfilov, Daniil Dzenhaliou, Caglar Gulcehre, Maksym Andriushchenko, Ameya Prabhu, Jonas Geiping

AI control protocols serve as a defense mechanism to stop untrusted LLM agents from causing harm in autonomous settings. Prior work treats this as a security problem, stress testing with exploits that use the deployment context to subtly complete harmful side tasks, such as backdoor insertion. In practice, most AI control protocols are fundamentally based on LLM monitors, which can become a central point of failure. We study \textit{adaptive} attacks by an untrusted model that knows the protocol and the monitor model, which is plausible if the untrusted model was trained with a later knowledge cutoff or can search for this information autonomously. We instantiate a simple adaptive attack vector by which the attacker embeds known or zero-shot prompt injections in the model outputs. Using this tactic, frontier models consistently evade diverse monitors and complete malicious tasks on two main AI control benchmarks. The attack works universally against current protocols that rely on a monitor. Furthermore, the recent Defer-to-Resample protocol even backfires, as its resampling amplifies the prompt injection and effectively reframes it as a best-of-$n$ attack. In general, adaptive attacks on monitor models represent a major blind spot in current control protocols and should become a standard component of evaluations for future AI control mechanisms.

Deep Learning · Generative Models and Autoencoders

Peiran Li, Jiawei Wang, Haoran Zhang, Xiaodan Shi, Noboru Koshizuka, Chihiro Shimizu, Renhe Jiang

The importance of mobile phone GPS trajectory data is widely recognized across many fields, yet the use of real data is often hindered by privacy concerns, limited accessibility, and high acquisition costs. As a result, generating pseudo–GPS trajectory data has become an active area of research. Recent diffusion-based approaches have achieved strong fidelity but remain limited in spatial scale (small urban areas), transportation-mode diversity, and efficiency (requiring numerous sampling steps). To address these challenges, we introduce TrajFlow, which to the best of our knowledge is the first flow-matching-based generative model for GPS trajectory generation. TrajFlow leverages the flow-matching paradigm to improve robustness and efficiency across multiple geospatial scales, and incorporates a trajectory harmonization \& reconstruction strategy to jointly address scalability, diversity, and efficiency. Using a nationwide mobile phone GPS dataset with millions of trajectories across Japan, we show that TrajFlow or its variants consistently outperform diffusion-based and deep generative baselines at urban, metropolitan, and nationwide levels. As the first nationwide, multi-scale GPS trajectory generation model, TrajFlow demonstrates strong potential to support inter-region urban planning, traffic management, and disaster response, thereby advancing the resilience and intelligence of future mobility systems.

Applications · Health

Philip Müller, Friederike Jungmann, Georgios Kaissis, Daniel Rueckert

Visual Question Answering (VQA) enables targeted and context-dependent analysis of medical images, such as chest X-rays (CXRs). However, existing VQA datasets for CXRs are typically constrained by simplistic and brief answer formats, lacking localization annotations (e.g., bounding boxes) and structured tags (e.g., region or radiological finding/disease tags). To address these limitations, we introduce MIMIC-Ext-CXR-QBA (abbr. CXR-QBA), a large-scale CXR VQA dataset derived from MIMIC-CXR, comprising 42 million QA-pairs with multi-granular, multi-part answers, detailed bounding boxes, and structured tags. We automatically generated our VQA dataset from scene graphs (also made available), which we constructed using LLM-based information extraction from radiology reports. After automatic quality assessment, we identified 31M pre-training and 7.5M fine-tuning grade QA-pairs, providing the largest and most sophisticated VQA dataset for CXRs to date. Tools for using our dataset and the construction pipeline are available at https://github.com/philip-mueller/mimic-ext-cxr-qba/ .

Social Aspects · Privacy-preserving Statistics and Machine Learning

Aviraj Newatia, Michael Cooper, Viet Nguyen, Rahul G. Krishnan

Training machine learning models requires the storage of large datasets, which often contain sensitive or private data. Storing data is associated with a number of potential risks which increase over time, such as database breaches and malicious adversaries. Machine unlearning is the study of methods to efficiently remove the influence of training data subsets from previously-trained models. Existing unlearning methods typically require direct access to the "forget set"---the data to be forgotten-and organisations must retain this data for unlearning rather than deleting it immediately upon request, increasing risks associated with the forget set. We introduce partially-blind unlearning---utilizing auxiliary information to unlearn without explicit access to the forget set. We also propose a practical framework Reload, a partially-blind method based on gradient optimization and structured weight sparsification to operationalize partially-blind unlearning. We show that Reload efficiently unlearns, approximating models retrained from scratch, and outperforms several forget set-dependent approaches. On language models, Reload unlearns entities using <0.025\% of the retain set and <7\% of model weights in <8 minutes on Llama2-7B. In the corrective case, Reload achieves unlearning even when only 10\% of corrupted data is identified.

Fang Wu, Xu Huang, Weihao Xuan, Zhiwei Zhang, Yijia Xiao, Frank Wan, Xiaomin Li, Bing Hu, Peng Xia, Jure Leskovec 等

Reinforcement learning from human feedback (RLHF) has emerged as the standard paradigm for aligning large language models with human preferences. However, reward-based methods grounded in the Bradley–Terry assumption struggle to capture the nontransitivity and heterogeneity of real-world preferences. To address this, recent studies have reframed alignment as a two-player Nash game, giving rise to Nash learning from human feedback (NLHF). While this perspective has inspired algorithms such as INPO, ONPO, and EGPO that offer strong theoretical and empirical guarantees, they remain fundamentally restricted to two-player interactions, introducing a single-opponent bias that fails to capture the full complexity of realistic preference structures. This work introduces Multiplayer Nash Preference Optimization (MNPO), a novel framework that generalizes NLHF to the multiplayer regime. It formulates alignment as an $n$-player game, where each policy competes against a population of opponents while being regularized toward a reference model. We demonstrate that MNPO inherits the equilibrium guarantees of two-player methods while enabling richer competitive dynamics and improved coverage of diverse preference structures. Comprehensive empirical evaluation shows that MNPO consistently outperforms existing NLHF baselines on instruction-following benchmarks, achieving superior alignment quality under heterogeneous annotator conditions and mixed-policy evaluation scenarios. Together, these results establish MNPO as a principled and scalable framework for aligning LLMs with complex, non-transitive human preferences. Code is available at~\url{https://github.com/smiles724/MNPO}.

Deep Learning · Graph Neural Networks

Alex Schutz, Victor-Alexandru Darvariu

Graph Neural Networks (GNNs) have achieved excellent results for modelling relational data in many supervised learning domains. However, much fewer works have explored their potential in Reinforcement Learning (RL) despite the ubiquity of practical problems defined over graphs. In this blog post, we discuss how GNNs can be effectively integrated in Deep RL frameworks, covering crucial design decisions and practical implementation concerns. In doing so, we hope to facilitate unlocking new capabilities for RL agents to reason in graph-structured environments with dynamic action spaces and varying input sizes.

Reinforcement Learning · Deep RL

Fang Wu, Xu Huang, Weihao Xuan, Zhiwei Zhang, Yijia Xiao, Frank Wan, Xiaomin Li, Bing Hu, Peng Xia, Jure Leskovec 等

Reinforcement learning from human feedback (RLHF) has emerged as the standard paradigm for aligning large language models with human preferences. However, reward-based methods grounded in the Bradley–Terry assumption struggle to capture the nontransitivity and heterogeneity of real-world preferences. To address this, recent studies have reframed alignment as a two-player Nash game, giving rise to Nash learning from human feedback (NLHF). While this perspective has inspired algorithms such as INPO, ONPO, and EGPO that offer strong theoretical and empirical guarantees, they remain fundamentally restricted to two-player interactions, introducing a single-opponent bias that fails to capture the full complexity of realistic preference structures. This work introduces Multiplayer Nash Preference Optimization (MNPO), a novel framework that generalizes NLHF to the multiplayer regime. It formulates alignment as an $n$-player game, where each policy competes against a population of opponents while being regularized toward a reference model. We demonstrate that MNPO inherits the equilibrium guarantees of two-player methods while enabling richer competitive dynamics and improved coverage of diverse preference structures. Comprehensive empirical evaluation shows that MNPO consistently outperforms existing NLHF baselines on instruction-following benchmarks, achieving superior alignment quality under heterogeneous annotator conditions and mixed-policy evaluation scenarios. Together, these results establish MNPO as a principled and scalable framework for aligning LLMs with complex, non-transitive human preferences. Code is available at~\url{https://github.com/smiles724/MNPO}.

Deep Learning · Attention Mechanisms

Songtao Liu, Hongwu Peng, Zhiwei Zhang, Zhengyu Chen, Yue Guo

Long-context inference in large language models is bottlenecked by Key--Value (KV) cache loading during the decoding stage, where the sequential nature of generation requires repeatedly transferring the KV cache from off-chip High-Bandwidth Memory (HBM) to on-chip Static Random-Access Memory (SRAM) at each step. While Multi-Head Latent Attention (MLA) significantly reduces the total KV cache size, it suffers from a sharding bottleneck during distributed decoding via Tensor Parallelism (TP). Since its single latent head cannot be partitioned, each device is forced to redundantly load the complete KV cache for every token, consuming excessive memory traffic and diminishing TP benefits like weight sharding. In this work, we propose Multi-Head Low-Rank Attention (MLRA), which enables partitionable latent states for efficient 4-way TP decoding. Extensive experiments show that MLRA achieves state-of-the-art perplexity and downstream task performance, while also delivering a 2.8$\times$ decoding speedup over MLA. Code is available at https://github.com/SongtaoLiu0823/MLRA. Pretrained weights, along with the training and evaluation data, are available at https://huggingface.co/Soughing/MLRA.

General Machine Learning · Transfer, Multitask and Meta-learning

Jackson Harmon, Andreas Hochlehnert, Matthias Bethge, Ameya Prabhu

Scaled post‑training now drives many of the largest capability gains in language models (LMs), yet its effect on pretrained knowledge remains poorly understood. Not all forgetting is equal: Forgetting one fact (e.g., a U.S. president or an API call) does not “average out” when recalling another. Hence, we propose a sample-wise paradigm to measure what is forgotten and when backward transfer occurs. Our metric counts 1→0 transitions (correct before post‑training, incorrect after) to quantify forgetting and 0→1 transitions to quantify backward transfer. Traditional task averages conflate these effects and obscure large changes. For multiple‑choice benchmarks, we add chance‑adjusted variants that subtract the expected contribution of random guessing from pre‑ and post‑training accuracies. We apply this framework across post‑training stages, model sizes, and data scales. Our large‑scale analysis across nearly 30 model pairs and 100 sub-benchmarks with up to 32,768 generated tokens per sample shows that: (1) Domain-continual pretraining induces moderate forgetting with low-to-moderate backward transfer; (2) RL/SFT post-training applied to base models and instruction tuning yields moderate-to-large backward transfer on math and logic with overall low-to-moderate forgetting; (3) Applying RL/SFT to instruction‑tuned models is sensitive on data scale: at small scales, both forgetting and backward transfer are small; at larger scales, effects are mixed and warrant further study with better controls; (4) Model merging does not reliably mitigate forgetting. Overall, our framework offers a practical yardstick for mapping how post‑training alters pretrained knowledge at scale -- enabling progress towards generally capable AI systems.

Reinforcement Learning · Deep RL

Timo Klein, Thomas Lang, Andrii Shkabrii, Alexander Sturm, Kevin Sidak, Lukas Miklautz, Claudia Plant, Yllka Velaj, Sebastian Tschiatschek

The exponential volume growth of hyperbolic geometry can embed the hierarchical relationships between states in reinforcement learning (RL) with far less distortion than Euclidean space. However, hyperbolic deep RL faces severe optimization challenges, and formal analysis of why optimization fails is lacking. We identify key factors that determine the success and failure of training hyperbolic deep RL agents. By analyzing the gradients of core operations in the Poincaré Ball and Hyperboloid models of hyperbolic geometry, we show that large-norm embeddings destabilize gradient-based training, leading to trust-region violations in proximal policy optimization (PPO). Based on these insights, we introduce Hyper++, a new hyperbolic deep RL agent that consists of three components: (1) feature regularization guaranteeing bounded norms while avoiding the curse of dimensionality from clipping; (2) a categorical value loss for stable critic training; and (3) a more optimization-friendly formulation of hyperbolic network layers. On ProcGen, we show that Hyper++ guarantees stable learning, outperforms prior hyperbolic agents, and reduces wall-clock time by approximately 30%. On Atari-5 with Double DQN, Hyper++ strongly outperforms Euclidean and hyperbolic baselines. We release our code at https://github.com/Probabilistic-and-Interactive-ML/hyper-rl.

Theory · Learning Theory

Jingyang Lyu, Kangjie Zhou, Yiqiao Zhong

Classification with imbalanced data is a common challenge in machine learning, where minority classes form only a small fraction of the training samples. Classical theory, relying on large-sample asymptotics and finite-sample corrections, is often ineffective in high dimensions, leaving many overfitting phenomena unexplained. In this paper, we develop a statistical theory for high-dimensional imbalanced linear classification, showing that dimensionality induces truncation or skewing effects on the logit distribution, which we characterize via a variational problem. For linearly separable Gaussian mixtures, logits follow $\\mathsf{N}(0,1)$ on the test set but converge to $\\max\\{\\kappa,\\mathsf{N}(0,1)\\}$ on the training set---a pervasive phenomenon we confirm on tabular, image, and text data. This phenomenon explains why the minority class is more severely affected by overfitting. We further show that margin rebalancing mitigates minority accuracy drop and provide theoretical insights into calibration and uncertainty quantification.

Applications · Robotics

Xupeng Zhu, Yu Qi, Yizhe Zhu, Robin Walters, Robert Platt

Multi-task manipulation policy often builds on transformer's ability to jointly process language instructions and 3D observations in a shared embedding space. However, real-world tasks frequently require robots to generalize to novel 3D object poses. Policies based on shared embedding break geometric consistency and struggle in 3D generation. To address this issue, we propose EquAct, which is theoretically guaranteed to generalize to novel 3D scene transformations by leveraging SE(3) equivariance shared across both language, observations, and action. EquAct makes two key contributions: (1) an efficient SE(3)-equivariant point cloud-based U-net with spherical Fourier features for policy reasoning, and (2) SE(3)-invariant Feature-wise Linear Modulation (iFiLM) layers for language conditioning. Finally, EquAct demonstrates strong spatial generalization ability and achieves state-of-the-art across $18$ RLBench tasks with both SE(3) and SE(2) scene perturbations, different amounts of training data, and on $4$ physical tasks.

Optimization · Large Scale, Parallel and Distributed

Tao Bu, Qiangang Wang, Bowen Zeng, Hanwen Sun, Yunpeng Huang, Chun Cao, Jingwei Xu

Transformer-based large language models (LLMs) have achieved remarkable success, yet their standard softmax-operator-based attention mechanism incurs quadratic computation and memory costs with respect to sequence length, posing a major bottleneck for long-context training. Prior work tackles this challenge along two directions: (1) kernel-level optimizations, which accelerate dense and sparse attention operators; and (2) module-level strategies, often referred to as distributed attention or context parallel training, which scale attention across multiple devices. However, systematic evaluation still remains limited: operator-level comparisons are often incomplete, while context parallel strategies are typically framework-specific, with unclear performance analysis across contexts. To address these gaps, we propose a unified benchmark that integrates representative attention kernels and context parallel mechanisms with a modular and extensible interface for evaluation. The benchmark evaluates methods along two critical dimensions: (1) attention mask patterns, which strongly affect efficiency, scalability, and usability, and (2) sequence length and distributed scale, which determine performance under extreme long-context training. Through comprehensive experiments on the cluster of up to 96 GPUs, our benchmark enables reproducible comparisons, highlights method-specific trade-offs, and provides practical guidance for designing and deploying attention mechanisms in long-context LLM training.

Dunjie Lu, Yiheng Xu, Junli Wang, Haoyuan Wu, Xinyuan Wang, Zekun Wang, Junlin Yang, Hongjin SU, Jixuan Chen, Junda Chen 等

Training computer-use agents requires massive amounts of GUI interaction data, but manually annotating action trajectories at scale is prohibitively expensive. We present VideoAgentTrek, a scalable pipeline that automatically mines training data from publicly available screen-recorded videos, eliminating the need for manual annotation. Our approach addresses a key challenge: raw videos contain implicit demonstrations but lack explicit action labels. To solve this, we develop Video2Action, an inverse dynamics module (IDM) with two components: (1) a video grounding model that detects and localizes GUI actions with precise temporal boundaries, and (2) an action-content recognizer that extracts structured parameters like click coordinates and typed text. Applied to 39,000 YouTube tutorial videos, our pipeline generates 1.52 million interaction steps. We leverage this data through continued pretraining followed by supervised fine-tuning. On OSWorld-Verified, our approach improves task success rates from 9.3% (SFT-only baseline) to 15.8%, a 70% relative improvement. On AgentNetBench, step accuracy increases from 64.1% to 69.3%. Our results demonstrate that passive internet videos can be transformed into high-quality supervision for computer-use agents, providing a scalable alternative to expensive manual annotation.

General Machine Learning · Everything Else

Qingyang Hao, Wenbo Liao, Bingyi Jing, Hongxin Wei

Selecting high-quality candidates from large-scale datasets is critically important in resource-constrained applications such as drug discovery, precision medicine, and the alignment of large language models. While conformal selection methods offer a rigorous solution with False Discovery Rate (FDR) control, their applicability is confined to single-threshold scenarios (i.e., y > c) and overlooks practical needs for multi-condition selection, such as conjunctive or disjunctive conditions. In this work, we propose the Multi-Condition Conformal Selection (MCCS) algorithm, which extends conformal selection to scenarios with multiple conditions. In particular, we introduce a novel nonconformity score with regional monotonicity for conjunctive conditions and a global Benjamini–Hochberg (BH) procedure for disjunctive conditions, thereby establishing finite-sample FDR control with theoretical guarantees. The integration of these components enables the proposed method to achieve rigorous FDR-controlled selection in various multi-condition environments. Extensive experiments validate the superiority of MCCS over baselines, its generalizability across diverse condition combinations, different real-world modalities, and multi-task scalability.

Liangcai Su, Zhen Zhang, Guangyu Li, Zhuo Chen, Chenxi Wang, Maojia Song, Xinyu Wang, Kuan Li, Jialong Wu, Xuanzhong Chen 等

Large language models (LLMs) have evolved into agentic systems capable of autonomous tool use and multi-step reasoning for complex problem-solving. However, post-training approaches building upon general-purpose foundation models consistently underperform in agentic tasks, particularly in open-source implementations. We identify the root cause: the absence of robust agentic foundation models forces models during post-training to simultaneously learn diverse agentic behaviors while aligning them to expert demonstrations, thereby creating fundamental optimization tensions. To this end, we are the first to propose incorporating Agentic Continual Pre-training (Agentic CPT) into the deep research agents training pipeline to build powerful agentic foundational models. Based on this approach, we develop a deep research agent model named AgentFounder. We evaluate our AgentFounder-30B on 10 benchmarks and achieve state-of-the-art performance while retains strong tool-use ability, notably 39.9% on BrowseComp-en, 43.3% on BrowseComp-zh, and 31.5% Pass@1 on HLE.

Applications · Language, Speech and Dialog

Lingjie Jiang, Shaohan Huang, xun wu, Yixia Li, Guanhua Chen, Dongdong Zhang, Furu Wei

Multimodal large language models (MLLMs) have significantly advanced the integration of visual and textual understanding. However, their ability to generate code from multimodal inputs remains limited. In this work, we introduce VisCodex, a unified framework that seamlessly merges vision and coding language models to empower MLLMs with strong multimodal code generation abilities. Leveraging a task vector-based model merging technique, we integrate a state-of-the-art coding LLM into a strong vision-language backbone, while preserving both visual comprehension and advanced coding skills. To support training and evaluation, we introduce the Multimodal Coding Dataset (MCD), a large-scale and diverse collection of 598k samples, including high-quality HTML code, chart image-code pairs, image-augmented StackOverflow QA, and algorithmic problems. Furthermore, we propose InfiBench-V, a novel and challenging benchmark specifically designed to assess models on visually-rich, real-world programming questions that demand a nuanced understanding of both textual and visual contexts. Extensive experiments show that VisCodex achieves state-of-the-art performance among open-source MLLMs and approaches proprietary models like GPT-4o, highlighting the effectiveness of our model merging strategy and new datasets.