We introduce STEERINGSAFETY, a benchmark for evaluating representation steering methods across nine safety perspectives spanning 18 datasets. While prior work highlights general capabilities of representation steering, we focus on safety perspectives including bias, harmfulness, hallucination, social behaviors, reasoning, epistemic integrity, and normative judgment. Our benchmark provides modularized building blocks for state-of-the-art steering methods, enabling unified implementation of DIM, ACE, CAA, PCA, and LAT with recent enhancements like conditional steering. Results on Gemma-2-2B, Llama-3.1-8B, and Qwen-2.5-7B reveal that strong steering performance depends critically on pairing of method, model, and specific perspective. For instance, DIM shows consistent effectiveness, but all methods exhibit substantial entanglement - where improving effectiveness on one perspective changes performance in other safety perspectives. Social behaviors show highest vulnerability (reaching degradation as high as 76%), jailbreaking often compromises normative judgment such as commonsense morality (degradation up to 26%), and hallucination steering unpredictably shifts political views, from 21% shifts right to 19% shifts to the political left. Our findings underscore the critical need for understanding steering methods from various safety angles.
论文检索
输入标题、作者或关键词，从 13,033 篇学术成果中精准定位
Sparse sketches such as the sparse Johnson–Lindenstrauss transform are a core primitive in randomized numerical linear algebra because they leverage random sparsity to reduce the arithmetic cost of sketching, while still offering strong approximation guarantees. Their random sparsity, however, is at odds with efficient implementations on modern GPUs, since it leads to irregular memory access patterns that degrade memory bandwidth utilization. Motivated by this tension, we pursue a sketch–kernel co-design approach: we design a new family of sparse sketches, BlockPerm-SJLT, whose sparsity structure is chosen to enable FlashSketch, a corresponding optimized CUDA kernel that implements these sketches efficiently. The design of BlockPerm-SJLT introduces a tunable parameter that explicitly trades off the tension between GPU-efficiency and sketching robustness. We provide theoretical guarantees for BlockPerm-SJLT under the oblivious subspace embedding (OSE) framework, and also analyze the effect of the tunable parameter on sketching quality. We empirically evaluate FlashSketch on standard RandNLA benchmarks, as well as an end-to-end ML data attribution pipeline called GraSS. FlashSketch pushes the Pareto frontier of sketching quality versus speed, across a range of regimes and tasks, and achieves a global geomean speedup of roughly $1.7 \times$ over the prior state-of-the-art GPU sketches.
Deep Learning · Theory
Many real-world datasets contain hidden structure that cannot be detected by simple linear correlations between input features. For example, latent factors may influence the data in a coordinated way, even though their effect is invisible to covariance-based methods such as PCA. In practice, nonlinear neural networks often succeed in extracting such hidden structure in unsupervised and self-supervised learning. However, constructing a minimal high-dimensional model where this advantage can be rigorously analyzed has remained an open theoretical challenge. We introduce a tractable high-dimensional spiked model with two latent factors: one visible to covariance, and one statistically dependent yet uncorrelated, appearing only in higher-order moments. PCA and linear autoencoders fail to recover the latter, while a minimal nonlinear autoencoder provably extracts both. We analyze both the population risk, and empirical risk minimization. Our model also provides a tractable example where self-supervised test loss is poorly aligned with representation quality: nonlinear autoencoders recover latent structure that linear methods miss, even though their reconstruction loss is higher.
While diffusion models have emerged as a powerful class of generative models, their learning dynamics remain poorly understood. We address this issue first by empirically showing that standard diffusion models trained on natural images exhibit a simplicity bias, learning simple, pair-wise input statistics before specializing to higher-order correlations. We reproduce this behaviour in simple denoisers trained on a minimal data model, the mixed cumulant model, where we precisely control both pair-wise and higher-order correlations of the inputs. We identify a scalar invariant of the model that governs the sample complexity of learning pair-wise and higher-order correlations that we call the _diffusion information exponent_, in analogy to related invariants in different learning paradigms. Using this invariant, we prove that the denoiser learns simple, pair-wise statistics of the inputs at linear sample complexity, while more complex higher-order statistics, such as the fourth cumulant, require at least cubic sample complexity. We also prove that the sample complexity of learning the fourth cumulant is linear if pair-wise and higher-order statistics share a correlated latent structure. Our work describes a key mechanism for how diffusion models can learn distributions of increasing complexity.
Deep Learning · Large Language Models
Test-time compute (TTC) has become an increasingly prominent paradigm for enhancing large language models (LLMs). Despite the empirical success of methods such as best-of-$n$ (BoN) sampling and sequential revision, their fundamental limits remain unclear. We address this gap by analyzing a mixture-of-reference policy model and proving that standard BoN is inherently suboptimal. To move closer to the optimal frontier, we study reward-filtered sequential inference, a simple procedure that selectively incorporates only high-reward generations into the context. This mechanism concentrates computation on superior policy candidates and suppresses inferior ones. On the theoretical side, we show that reward-filtered sequential inference yields strictly stronger guarantees than standard TTC paradigms. On the empirical side, we evaluate such an inference strategy across diverse benchmarks and observe consistent improvements over widely used approaches, demonstrating the practical effectiveness of our framework.
Applications · Chemistry, Physics, and Earth Sciences
Generating long-horizon, all-atom molecular dynamics (MD) is difficult due to error accumulation in time-domain autoregressive models (causing drift) and fixed step-size constraints on temporal resolution. We propose **BioDynaSpec**, which reformulates protein dynamics as spatio-spectral generation: **Independent Windowed Fourier Decomposition (IWFD)** decomposes trajectories into independent windowed frequency representations, and a generator combines low-to-high frequency autoregression with diffusion denoising to reconstruct continuous motion. We introduce **Inter-Residue Frequency Coupling (IRFC)** bias, a learnable Gaussian distance bias in attention that injects a resonance-inspired structural prior to stabilize training and improve cross-residue, cross-frequency consistency. On ATLAS, BioDynaSpec improves 250-frame trajectory generation with $R_{250}=1.509$ Å (where $R_s$ denotes the mean per-frame C$\alpha$-RMSE over the first $s$ frames after alignment), reducing error by 60.4\% vs. MDGEN and 57.2\% vs. ProAR, and achieves the best PCA-2D displacement-profile correlation and stepwise distribution matching. For equilibrium conformational sampling, it achieves Root Mean $W_2=1.31$, MD PCA $W_2=0.90$, and Joint PCA $W_2=1.19$ (50.03\%, 35.25\%, and 47.58\% lower than the next best), while ablations show removing IRFC severely degrades RMSE/MAE and correlation.
Deep Learning · Large Language Models
Although reinforcement learning (RL) enhances the reasoning capabilities of large language models (LLMs), it is primarily learned from the model's self-generated distribution, limiting its ability to acquire reasoning skills beyond its initial knowledge. To overcome this, we propose a Difficulty-Aware Learning Strategy Allocation (DALSA) framework, which adaptively assigns appropriate learning strategies to samples based on their difficulty signals. DALSA is built on the key insight that samples beyond models' knowledge scope are better addressed through supervised fine-tuning (SFT), while those within the boundary but insufficiently mastered benefit more from RL, and well-learned samples are discarded to avoid redundant updates. To realize this principle, we extract a series of difficulty-aware training characteristics and employ a learnable strategy allocator to dynamically determine the optimal learning strategy for each sample based on its training dynamics. The allocator and the LLM are alternately optimized, enabling adaptive strategy allocation. Furthermore, two regularization techniques, anti-curriculum weighting and adversarial label smoothing, are integrated to alleviate the inherent limitations of RL and SFT, backed by comprehensive theoretical analyses. Extensive experiments on ten LLMs ranging from 1.5B to 70B across various tasks indicate that DALSA consistently outperforms baselines under both full and parameter-efficient fine-tuning settings.
Deep Learning · Large Language Models
Fine-tuned Large Language Models (LLMs) are vulnerable to backdoor attacks through data poisoning, yet the internal mechanisms governing these attacks remain a black box. Previous research on interpretability for LLM safety tends to focus on alignment, jailbreak, and hallucination, but overlooks backdoor mechanisms, making it difficult to understand and fully eliminate the backdoor threat. In this paper, aiming to bridge this gap, we explore the interpretable mechanisms of LLM backdoors through Backdoor Attribution (BkdAttr), a tripartite causal analysis framework. We first introduce the Backdoor Probe that proves the existence of learnable backdoor features encoded within the representations. Building on this insight, we further develop Backdoor Attention Head Attribution (BAHA), efficiently pinpointing the specific attention heads responsible for processing these features. Our primary experiments reveals these heads are relatively sparse; ablating a minimal \textbf{$\sim$ 3%} of total heads is sufficient to reduce the Attack Success Rate (ASR) by \textbf{over 90%}. More importantly, we further employ these findings to construct the Backdoor Vector derived from these attributed heads as a master controller for the backdoor. Through only \textbf{1-point} intervention on \textbf{single} representation, the vector can either boost ASR up to \textbf{$\sim$ 100% $\uparrow$} on clean inputs, or completely neutralize backdoor, suppressing ASR down to \textbf{ $\sim$ 0%} on triggered inputs. In conclusion, our work pioneers the exploration of mechanistic interpretability in LLM backdoors, demonstrating a powerful method for backdoor control and revealing actionable insights for the community.
Deep Learning · Generative Models and Autoencoders
In the domain of text-to-video (T2V) generation, reliably synthesizing compositional content involving multiple subjects with intricate relations is still underexplored. The main challenges are twofold: 1) Subject presence, where not all subjects can be presented in the video; 2) Inter-subject relations, where the interaction and spatial relationship between subjects are misaligned. Existing methods adopt techniques, such as inference-time latent optimization or layout control, which fail to address both issues simultaneously. To tackle these problems, we propose \textbf{Comp-Attn}, a composition-aware cross-attention variant that follows a \textbf{``Present-and-Align” }paradigm: it decouples the two challenges by enforcing subject presence at the condition level and achieving relational alignment at the attention-distribution level. Specifically, 1) We introduce Subject-aware Condition Interpolation (SCI) to reinforce subject-specific conditions and ensure each subject's presence; 2) We propose Layout-forcing Attention Modulation (LAM), which dynamically enforces the attention distribution to align with the relational layout of multiple subjects. Comp-Attn can be seamlessly integrated into various T2V baselines in a training-free manner, boosting T2V-CompBench scores by 15.7\% and 11.7\% on Wan2.1-T2V-14B and Wan2.2-T2V-A14B with only a 5\% increase in inference time. Meanwhile, it also achieves strong performance on VBench and T2I-CompBench, demonstrating its scalability in general video generation and compositional text-to-image (T2I) tasks.
We introduce Unsupervised Partner Design (UPD), a population-free multi-agent reinforcement learning method for robust ad-hoc teamwork. UPD generates training partners on-the-fly and selects them adaptively based on a learnability criterion, removing the need for pre-trained partner populations or manual parameter tuning. We show that this simple mechanism enables effective partner diversity and can be extended to joint partner-environment selection when a procedural level generator is available. Across Level-Based Foraging, Overcooked-AI, and the Overcooked Generalisation Challenge, UPD consistently outperforms both population-based and population-free baselines. In a human-AI user study, agents trained with UPD achieve higher returns and are rated as more adaptive, more human-like, and less frustrating than existing approaches.
Applications · Computer Vision
Existing video object removal methods predominantly rely on diffusion models following a noise-to-data paradigm, where generation starts from uninformative Gaussian noise. This approach discards the rich structural and contextual priors present in the original input video. Consequently, such methods often lack sufficient guidance, leading to incomplete object erasure or the synthesis of implausible content that conflicts with the scene's physical logic. In this paper, we reformulate video object removal as a video-to-video translation task via a stochastic bridge model. Unlike noise-initialized methods, our framework establishes a direct stochastic path from the source video (with objects) to the target video (objects removed). This bridge formulation effectively leverages the input video as a strong structural prior, guiding the model to perform precise removal while ensuring that the filled regions are logically consistent with the surrounding environment. To address the trade-off where strong bridge priors hinder the removal of large objects, we propose a novel adaptive mask modulation strategy. This mechanism dynamically modulates input embeddings based on mask characteristics, balancing background fidelity with generative flexibility. Extensive experiments demonstrate that our approach significantly outperforms existing methods in both visual quality and temporal consistency. The project page is https://bridgeremoval.github.io/.
Deep Learning · Large Language Models
Real-world user requests to LLM agents are often underspecified. Agents must interact to acquire missing information and make correct downstream decisions. However, current multi-turn GRPO-based methods often rely on trajectory-level reward computation, which leads to credit assignment problems and insufficient advantage signals within rollout groups. A feasible approach is to identify valuable interaction turns at a fine granularity to drive more targeted learning. To address this, we introduce InfoPO, which frames multi-turn interaction as a process of active uncertainty reduction and computes an information-gain reward that credits turns whose feedback measurably changes the agent’s subsequent action distribution compared to a masked-feedback counterfactual. It then combines this signal with task outcomes via an adaptive variance-gated fusion to identify information importance while maintaining task oriented goal direction. Across diverse tasks including intent clarification, collaborative coding, and tool-augmented decision making, InfoPO consistently outperforms prompting and multi-turn RL baselines. It also demonstrates robustness under user simulator shifts and generalizes effectively to environment interactive tasks. Overall, InfoPO provides a principled and scalable mechanism for optimizing complex agent user collaboration.
Deep Learning · Large Language Models
Large Language Models (LLMs) fundamentally suffer from representation collapse, a bottleneck that severely degrades performance in long contexts. We identify that existing approaches risk drifting into one of two pathological extremes: Homogenization Collapse (e.g., attention sinks causing rank deficiency) and Isolation Collapse (e.g., local attention causing context disconnection). Through spectral analysis of attention dynamics, we derive an intrinsic trade-off between Mixing Efficiency (spectral gap) and Information Capacity (effective rank), revealing that standard mechanisms struggle to maximize both simultaneously. To resolve this dilemma, we propose the Topologically Regularized Side-Path (TRSP), a non-invasive architectural intervention designed to achieve spectral balance. TRSP employs a parameter-free Triangular Box mechanism scaled by a lightweight, length-aware gate to explicitly regularize the token interaction topology. By integrating proximal coupling to preserve the effective rank and distal propagation to guarantee the spectral gap, this design ensures a geometrically healthy state without altering the core attention mechanism. Experiments yield significant performance improvements across general capabilities and long-context benchmarks. Notably, on the NoLiMa extrapolation benchmark at 8$\times$ the training length, TRSP surpasses strong baselines like the Differential Transformer and Gated Attention by approximately 30\% and 50\%, respectively.
Applications · Computer Vision
Scene text editing aims to modify text in a target region of an image while preserving its background style and texture. Existing methods rely solely on image background information while neglecting the visual details of target regions, which discards stylistic features in the original text and essentially degrades the task to text rendering. Moreover, the conditions imposed by pre-trained glyph encoder limit the scope of editable text. To address these issues, this paper proposes a self-prompting scene text editing method, which constructs style and glyph prompts directly from the original image without additional style or glyph encoders. We employ a two-stage training strategy, where the diffusion transformer is first trained on large-scale self-supervised datasets and subsequently refined with a small set of paired images. By leveraging the in-context learning capability of FLUX-Fill, it achieves open-vocabulary and style-consistent text editing. Experimental results on various languages demonstrate that our method achieves the state-of-the-art performance in both text accuracy and style consistency.
Concept erasure has emerged as a promising approach to mitigate undesired or unsafe content in diffusion models, yet existing methods still face significant limitations. While training-based methods are effective, their high computational cost limits scalability. Editing-based methods are more efficient and deployment-friendly, yet they struggle to simultaneously achieve precise concept erasure and preserve overall generative capacity. We identify this core limitation of the editing-based methods as reliance on additive parameter updates. Our empirical analysis reveals that concept semantics primarily depend on *neuron direction* rather than *neuron magnitude*, while overall generative capacity relies on the *angular geometry* of neurons. As additive updates inherently entangle direction, magnitude, and angular geometry, they inevitably introduce unintended interference between concept erasure and overall generation performance. To address this, we propose **Orthogonal Concept Erasure (OCE)**, which reformulates editing-based erasure as multiplicative parameter updates from a geometric perspective. Specifically, OCE applies layer-wise orthogonal transformations derived from a closed-form solution to the parameters, enabling precise concept erasure while preserving the neuron magnitude and angular geometry. Furthermore, to address conflicting constraints in multi-concept erasure, OCE introduces a subspace-level objective with structured subspace manipulation, yielding a more effective and scalable erasure. Extensive experiments on single- and multi-concept erasure demonstrate that OCE outperforms existing methods in concept erasure and non-target preservation, erasing up to 100 concepts in 4.3 s.
Applications · Computer Vision
LLM-conditioned segmentation has recently advanced rapidly by coupling large language models with iterative mask generation frameworks. However, we identify a persistent failure mode in current propose-then-select pipelines. Although high-quality mask candidates are often generated, the final prediction may fail to match the given linguistic condition. This failure arises because language semantics are typically used as static prompts or post-hoc matching signals, rather than participating in the iterative mask generation process. Through systematic analysis, we show that many errors stem from semantic misalignment rather than poor mask quality. To address this issue, we propose FlowSeg, which introduces dynamic semantic guidance via a bidirectional semantic flow between intermediate decoding states and LLM-derived condition embeddings throughout the generation process. Language conditions actively guide mask refinement at each stage, while condition embeddings are progressively updated by emerging visual evidence. This design yields semantically grounded mask representations and visually aligned language conditions, enabling more reliable matching. We further incorporate a lightweight boundary-aware refinement to selectively enhance uncertain regions without perturbing confident interiors. Extensive experiments on referring expression segmentation and reasoning segmentation tasks demonstrate that FlowSeg consistently improves language–mask alignment and achieves state-of-the-art performance.
Social Aspects · Fairness
With the on-going integration of machine learning systems into the everyday social life of millions the notion of fairness becomes an ever increasing priority in their development. Fairness notions commonly rely on protected attributes to assess potential biases. Here, the majority of literature focuses on discrete setups regarding both target and protected attributes. The literature on continuous attributes especially in conjunction with regression - we refer to this as *continuous fairness* - is scarce. A common strategy is iterative null-space projection which as of now has only been explored for linear models or embeddings such as obtained by a non-linear encoder. We improve on this by extending this to kernel induced feature spaces by means of the ``empirical feature space''. We theoretically derive this as a direct transformation of the kernel matrix yielding a model and fairness-score agnostic method applicable to continuous protected attributes. We demonstrate that our novel approach in conjunction with Support Vector Regression (SVR) provides competitive or improved performance across multiple datasets in comparisons to other contemporary methods.
Deep Learning · Large Language Models
Mixture-of-Experts Large Language Models (MoE-LLMs) achieve strong performance but incur substantial memory overhead due to massive expert parameters. Mixed-precision quantization mitigates this cost by allocating expert-wise bit-widths based on their importance, approaching the accuracy-memory Pareto frontier and enabling extreme low-bit quantization. However, existing methods rely on layer-wise importance estimation and overlook router shifts induced by quantization, resulting in suboptimal allocation and routing. In this work, we propose Global Expert-level Mixed-precision Quantization (GEMQ) to overcome these limitations via (1) a global linear-programming formulation that captures model-wide expert importance based on quantization error analysis, and (2) efficient router fine-tuning to adapt routing to quantized experts. These components are integrated into a progressive quantization framework that iteratively refines importance estimation and allocation. Experiments demonstrate that GEMQ significantly reduces memory and accelerates inference with minimal accuracy degradation.
Diffusion models trained on different, non-overlapping subsets of a dataset often produce strikingly similar outputs when given the same noise seed. We trace this consistency to a simple linear effect: the shared Gaussian statistics across splits already predict much of the generated images. To formalize this, we develop a random matrix theory (RMT) framework that quantifies how finite datasets shape the expectation and variance of the learned denoiser and sampling map in the linear setting. For expectations, sampling variability acts as a renormalization of the noise level through a self-consistent relation $\sigma^2\to\kappa(\sigma^2)$, explaining why limited data overshrink low-variance directions and pull samples toward the dataset mean. For fluctuations, our variance formulas reveal three key factors behind cross-split disagreement: \textit{anisotropy} across eigenmodes, \textit{inhomogeneity} across inputs, and overall scaling with dataset size. Extending deterministic-equivalence tools to fractional matrix powers further allows us to analyze entire sampling trajectories. The theory sharply predicts the behavior of linear diffusion models, and we validate its predictions on UNet and DiT architectures in their non-memorization regime, identifying where and how samples deviates across training data split. This provides a principled baseline for reproducibility in diffusion training, linking spectral properties of data to the stability of generative outputs.
Recent research has established that the impact of context in a vanilla transformer can be represented implicitly by forming a token-dependent, rank-1 patch to its MLP weights. This work extends that foundational theory to the diverse architectures of modern Large Language Models. We first demonstrate a precise, analytical solution for a Gemma-style transformer block, proving that the entire effect of a context can be perfectly mapped to rank-1 patches on its MLP weight matrices and a patch to the RMSNorm scale. We then generalize this result, providing a constructive proof and algorithm for multi-layer models. To unify these findings, we introduce a general framework centered on two core properties: input controllability and output controllability. We prove that a perfect implicit weight patch is possible for any MLP block where the inner function is input-controllable and the outer function is output-controllable. This provides a simpler and more powerful lens for understanding how transformer models transmute prompts into effective weights. This setup generalizes to a wide range of modern LLM architectures including gating, pre-/post-norm, mixture of experts and sequential/parallel transformer blocks.