Pre-trained Vision-Language Models (VLMs) such as CLIP achieve strong zero-shot generalization, but their performance degrades sharply under adversarial perturbations. Existing test-time adaptation methods typically rely on sample-level confidence heuristics, overlooking the intrinsic distributional structure of the data. This sample-centric approach limits robustness, as it fails to distinguish confident adversarial mispredictions from true semantic consistency. In this work, we observe that adversarial distortion is structurally brittle: while holistic representations are corrupted, semantic integrity is often preserved in the distribution of augmented views. Motivated by this insight, we propose $\texttt{RITA}$, a $\textbf{R}$obust test-t$\textbf{I}$me promp$\textbf{T}$ $\textbf{A}$daptation framework that shifts from sample-level estimates to distribution-level alignment. Specifically, $\texttt{RITA}$ employs optimal transport to align the distribution of augmented visual features with textual prototypes, mitigating adversarial outliers and rectifying cross-modal semantic misalignment. Furthermore, we introduce a dynamic cache to progressively accumulate reliable cues from the test stream for online refinement. Extensive experiments demonstrate that $\texttt{RITA}$ significantly improves adversarial robustness without compromising clean accuracy.
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
Deep Learning · Foundation Models
This position paper argues that the machine learning community should prioritize early-stage quality assurance in annotation pipelines over the prevailing practice of late-stage validation. Data quality bottlenecks increasingly limit foundation model improvement, yet quality assurance research focuses almost exclusively on validation methods rather than validation timing. *When* validation occurs—not merely *what* validation methods are employed—fundamentally determines both error rates and annotation costs. This temporal neglect is puzzling given the well-established "shift-left" principle from software engineering, where empirical studies demonstrate 4–100× cost multipliers for defects detected in later development stages (Boehm, 1981; Shull et al., 2002). Annotation pipelines, we argue, exhibit analogous dynamics: errors caught before annotation begins cost a fraction of those discovered after review cycles complete. We propose a taxonomy of three *QA trigger points*—pre-annotation (T₀), post-annotation (T₁), and post-review (T₂)—that decompose annotation workflows into discrete validation opportunities. A survey of 47 recent papers reveals that only 4% report when validation occurs, a striking gap given timing's demonstrated impact in adjacent fields. Without explicit attention to QA timing, the community risks optimizing validation methods while ignoring the structural variable that may matter most. We call on researchers to report QA timing configurations, on platform developers to expose timing as a first-class parameter, and on the community to conduct controlled experiments testing whether the shift-left principle transfers to annotation contexts.
Applications · Health / Medicine
Most existing CLIP-style medical vision--language pretraining methods rely on global or local alignment with substantial paired data. However, global alignment is easily dominated by non-diagnostic information, while local alignment fails to integrate key diagnostic evidence. As a result, learning reliable diagnostic representations becomes difficult, which limits their applicability in medical scenarios with limited paired data. To address this issue, we propose an LLM-Guided Diagnostic Evidence Alignment method (LGDEA), which shifts the pretraining objective toward evidence-level alignment that is more consistent with the medical diagnostic process. Specifically, we leverage LLMs to extract key diagnostic evidence from radiology reports and construct a shared diagnostic evidence space, enabling evidence-aware cross-modal alignment and allowing LGDEA to effectively exploit abundant unpaired medical images and reports, thereby substantially alleviating the reliance on paired data. Extensive experimental results demonstrate that our method achieves consistent and significant improvements on phrase grounding, image--text retrieval, and zero-shot classification, and even rivals pretraining methods that rely on substantial paired data.
Applications · Language, Speech and Dialog
Emotional expression in human speech is nuanced and compositional, often involving multiple, sometimes conflicting, affective cues that may diverge from linguistic content. In contrast, most expressive text-to-speech (TTS) systems enforce a single utterance-level emotion, collapsing affective diversity and suppressing mixed or text–emotion–misaligned expression. While activation steering via latent direction vectors offers a promising solution, it remains unclear whether emotion representations are linearly steerable in TTS, where steering should be applied within hybrid TTS architectures, and how such complex emotion behaviors should be evaluated. This paper presents the first systematic analysis of activation steering for emotional control in hybrid TTS models, introducing a quantitative, controllable steering framework, and multi-rater evaluation protocols that enable composable mixed-emotion synthesis and reliable text–emotion mismatch synthesis. Our results demonstrate, for the first time, that emotional prosody and expressive variability are primarily synthesized by the TTS language module instead of the flow-matching module, and also provide a lightweight steering approach for generating natural, human-like emotional speech.
Probabilistic Methods · Variational Inference
Real-world data is frequently noisy and ambiguous. In crowdsourcing, for example, human annotators may assign conflicting class labels to the same instances. Partial-label learning (PLL) addresses this challenge by training classifiers when each instance is associated with a set of candidate labels, only one of which is correct. While early PLL methods approximate the true label posterior, they are often computationally intensive. Recent deep learning approaches improve scalability but rely on surrogate losses and heuristic label refinement. We introduce a novel probabilistic framework that directly approximates the posterior distribution over true labels using amortized variational inference. Our method employs neural networks to predict variational parameters from input data, enabling efficient inference. This approach combines the expressiveness of deep learning with the rigor of probabilistic modeling, while remaining architecture-agnostic. Theoretical analysis and extensive experiments on synthetic and real-world datasets demonstrate that our method achieves state-of-the-art performance in both accuracy and efficiency.
Social Aspects · Safety
Large Audio–Language Models (LALMs) are becoming essential as a powerful multimodal backbone for real-world applications. However, recent studies show that audio inputs can more easily elicit harmful responses than text, exposing new risks toward deployment. While safety alignment has made initial advances in LLMs and Large Vision–Language Models (LVLMs), we find that vanilla adaptation of these approaches to LALMs faces two key limitations: 1) LLM-based steering fails under audio input due to the large distributional gap between activations, and 2) prompt-based defenses induce over-refusals on benign-speech queries. To address these challenges, we propose **S**afe-**A**blated **R**efusal **Steer**ing (SARSteer), an effective inference-time defense framework for LALMs. Specifically, SARSteer leverages text-derived refusal steering to enforce rejection without manipulating audio inputs and introduces decomposed safe-space ablation to mitigate over-refusal. Extensive experiments demonstrate that SARSteer significantly improves harmful-query refusal while preserving benign responses, establishing a principled step toward safety alignment in LALMs.
Social Aspects · Safety
In this position paper, we address the persistent gap between rapidly growing AI capabilities and lagging safety progress. Existing paradigms divide into "Make AI Safe", which applies post-hoc alignment and guardrails but remains brittle and reactive, and "Make Safe AI", which emphasizes intrinsic safety but struggles to address unforeseen risks in open-ended environments. We therefore propose safe-by-coevolution as a new formulation of the "Make Safe AI" paradigm, inspired by biological immunity, in which safety becomes a dynamic, adversarial, and ongoing learning process. To operationalize this vision, we introduce R$^2$AI---Resistant and Resilient AI---as a practical framework that unites resistance against known threats with resilience to unforeseen risks. R$^2$AI integrates fast and slow safe models, adversarial simulation and verification through a safety wind tunnel, and continual feedback loops that guide safety and capability to coevolve. We argue that this framework offers a scalable and proactive path to maintain continual safety in dynamic environments, addressing both near-term vulnerabilities and long-term existential risks as AI advances toward AGI and ASI.
Deep Learning · Graph Neural Networks
Graph Neural Networks (GNNs) are powerful tools for learning from graph-structured data, but their scalability is increasingly strained by the size of real-world graphs in domains like recommender systems, fraud detection, and molecular biology. Graph condensation—the task of generating a smaller synthetic graph that retains the performance of models trained on the original—has emerged as a promising solution. However, the dominant approach of gradient matching introduces a fundamental contradiction: it requires training on the full dataset to create the compressed version, thereby undermining the goal of efficiency. Worse still, these methods suffer from high computational overhead, poor generalization across GNN architectures, and brittle reliance on specific model configurations. Equally concerning is the community's reliance on misleading evaluation protocols such as node compression ratios, which fail to reflect true resource savings, condensation overhead, and illusory application to neural architecture search. These shortcomings are not incidental—they are systemic, and they obstruct meaningful progress. In this position paper, we argue that graph condensation, in its current form, needs a reset. We call for moving beyond full-dataset training and model-dependent design, and instead advocate for methods that are lightweight, architecture-agnostic, and practically deployable. By identifying key methodological flaws and outlining concrete research directions, we aim to reorient the field toward approaches that deliver on the true promise of condensation: efficient, generalizable, and usable GNN training at scale.
Reinforcement Learning · Everything Else
Safe reinforcement learning (RL) is a standard paradigm for safety-critical decision making. However, real-world safety constraints can be complex, subjective, and even hard to explicitly specify. Existing works on constraint inference rely on restrictive assumptions or extensive expert demonstrations, which is not realistic in many real-world applications. How to cheaply and reliably learn these constraints is the major challenge we focus on in this study. While inferring constraints from human preferences offers a data-efficient alternative, we identify the popular Bradley-Terry (BT) models fail to capture the asymmetric, heavy-tailed nature of safety costs, resulting in risk underestimation. It is still rare in the literature to understand the impacts of BT models on the downstream policy learning. To address the above knowledge gaps, we propose a novel approach namely Preference-based Constrained Reinforcement Learning (PbCRL). We introduce a novel dead zone mechanism into preference modeling and theoretically prove that it encourages heavy-tailed cost distributions, thereby achieving better constraint alignment. Additionally, we incorporate a Signal-to-Noise Ratio (SNR) loss to encourage exploration by cost variances, which is found to benefit policy learning. Further, two-stage training strategy are deployed to lower online labeling burdens while adaptively enhancing constraint satisfaction. Empirical results demonstrate that PbCRL achieves superior alignment with true safety requirements and outperforms the state-of-the-art baselines in terms of safety and reward. Our work explores a promising and effective way for constraint inference in Safe RL, which has great potential in a range of safety-critical applications.
Social Aspects · Safety
AI advancements have been significantly driven by a combination of foundation models and curiosity-driven learning aimed at increasing capability and adaptability. Within this landscape, open-endedness, where AI agents autonomously and indefinitely generate novel behaviors, representations, or solutions, has gained increasing interest. This has become relevant in the context of self-evolving agents and long-horizon discovery. This position paper argues that the defining properties of open-ended AI systems introduce a distinct and underexplored class of safety challenges, including loss of predictability, emergent misalignment, and difficulties in maintaining effective control as systems evolve beyond their initial design assumptions, that must be addressed preemptively. These challenges differ qualitatively from those associated with task-bounded or static models and are unlikely to be addressed by existing safety frameworks alone, which is why these risks must be examined proactively, before large-scale deployment. The paper outlines key challenges, discusses research opportunities, and calls for coordinated action to support the safe and responsible development of open-ended AI.
Applications · Everything Else
AI systems can strengthen democracy by supporting deliberation at scale by addressing cognitive, social, platform-design, and market-driven frictions, while preserving human agency. Unlike proposals such as liquid democracy that restructure representation through vote delegation, in this position paper, we argue that AI-assisted deliberation offers a more promising path by lowering barriers to meaningful engagement without substituting machine judgment for human choice. Drawing on evidence from online platforms and experimental research, we identify four guiding principles: preserving agency and autonomy, encouraging mutual respect, promoting equality and inclusiveness, and augmenting rather than substituting active citizenship. We also address critical challenges, including alignment, sycophancy, training bias, and over-reliance on AI systems. We call on the machine learning community to develop deliberation-focused AI systems evaluated not on engagement metrics but on their capacity to facilitate informed, representative, and friction-robust discourse.
Deep Learning · Large Language Models
In-context learning is fundamental to modern Large Language Models (LLMs); however, prevailing architectures impose a rigid and fixed contextual structure by assigning linear or constant positional indices. Drawing on Cognitive Load Theory (CLT), we argue that this uninformative structure increases extraneous cognitive load, consuming finite working memory capacity that should be allocated to deep reasoning and attention allocation. To address this, we propose RePo, a novel mechanism that reduces extraneous load via context re-positioning. Unlike standard approaches, RePo utilizes a differentiable module, $f_\phi$, to assign token positions that capture contextual dependencies, rather than replying on pre-defined order. By continually pre-training on the OLMo-2 1B \& 7B models, we demonstrate that RePo consistently enhances performance on tasks involving noisy contexts, structured data, and longer context length, while maintaining competitive performance on general short-context tasks. Detailed analysis reveals that RePo successfully allocate higher attention to distant but relevant information, assign positions in dense and non-linear space, and capture the intrinsic structure of the input context. We will open-source the code and model weights.
Deep Learning · Algorithms
Many modern AI systems are designed to operate under diverse, open-ended, use-cases. To help generalize deployed systems, developers rely on a reactive AI flywheel that observes emerging feedback from user behavior (errors) and patches the model accordingly. However, most flywheels ignore the broader context of these errors within the system's objectives, failing to preempt potential future edge cases, which leads to more unnecessary flywheel iterations. Also, it is statistically increasingly difficult to collect remaining errors due to the long-tail nature of open-world use-cases (Boneh and Hofri, 1997). This position paper argues that a *proactive test-driven flywheel* is required to address reactive flywheel's limitations and to approach a generalizable system. We advocate for creating a ``test space" to technically map feedback data to task objectives, evolving the flywheel from reactive to proactive. We augment our position by mathematically proving a proactive one achieves better long-term scaling with fewer iterations than the reactive flywheel.
Deep Learning · Self-Supervised Learning
Out-of-distribution (OOD) detection is critical for reliable deployment of vision models. Mahalanobis-based detectors remain strong baselines, yet their performance varies widely across modern pretrained representations, and it is unclear which properties of a feature space cause these methods to succeed or fail. We conduct a large-scale study across diverse foundation-model backbones and Mahalanobis variants. First, we show that Mahalanobis-style OOD detection is not universally reliable: performance is highly representation-dependent and can shift substantially with pretraining data and fine-tuning regimes. Second, we link this variability to in-distribution geometry and identify a two-term ID summary that consistently tracks Mahalanobis OOD behavior across detectors: within-class spectral structure and local intrinsic dimensionality. Finally, we treat normalization as a geometric control mechanism and introduce radially scaled $\ell_2$ normalization, $\phi_\beta(z)=z/\|z\|^\beta$, which preserves directions while contracting or expanding feature radii. Varying $\beta$ changes the radii while preserving directions, so the same quadratic detector sees a different ID geometry. We choose $\beta$ from ID-only geometry signals and typically outperform fixed normalization baselines.
Reinforcement Learning · Everything Else
One goal in reinforcement learning (RL) research is to understand general purpose sequential decision-making, using benchmark simulators as a proxy for learning in a deployment setting. When running experiments, however, the goal of achieving high performance in the simulator can mutate into focusing exclusively on solving the simulator. To achieve high scores researchers may adopt solutions exclusively meant for solving simulators, rather than learning while the agent is deployed outside of a simulator. Solving simulators is also worthy of investigation, but is a fundamentally different RL research question. *In this paper we argue that RL researchers need to distinguish between two uses cases of simulators: solving simulators and using simulators as a proxy for learning in deployment.* We first discuss how these two use-cases are importantly different, in terms of constraints on how the agent can use the simulator, which algorithms are appropriate and which evaluation metrics are appropriate. We then highlight several issues and misleading conclusions that can occur by not making the distinction between these two settings clear, supported with examples and simple experiments. This work is a call to the community to begin clearly distinguishing how they are using simulators in their work, hopefully sparking further discussion on which empirical practices work best in each setting.
Deep Learning · Large Language Models
We identify two empirical issues in large language model (LLM) training: (i) optimizer updates can have large spectral norms, potentially destabilizing training and degrading generalization; (ii) stochastic gradient noise can exhibit sparse spectral spikes, with a few dominant singular values much larger than the rest. We propose *SPECTRA*, a general framework addressing these by (i) *post*-spectral clipping of updates to enforce spectral-norm constraints (ii) optional *pre*-spectral clipping of gradients to suppress spectral noise spikes. We prove that post-clipping constitutes a Composite Frank-Wolfe method with spectral-norm constraints and weight regularization, recovering Frobenius and $\ell_{\infty}$-norm regularization with SGD-based and sign-based methods. We further analyze how pre-clipping mitigates sparse spectral spikes. We propose efficient soft spectral clipping via Newton-Schulz iterations, avoiding expensive SVD. Experiments on LLM pretraining show SPECTRA uniformly improves validation loss for various optimizers, including AdamW, Signum, and AdEMAMix, with the best-performing variants achieving state-of-the-art results. Models trained with SPECTRA exhibit smaller weight norms, confirming the link between spectral clipping and regularization.
Deep Learning · Large Language Models
Unified Foundation Models (UFMs), which support interleaved multimodal generation and understanding, have been proposed as a promising paradigm for reasoning about dynamic world states, yet it remains unclear whether the visual content they generate functions as grounded evidence for subsequent reasoning or merely as auxiliary output. Existing benchmarks largely evaluate generation and understanding as separate capabilities and do not test their functional dependence during reasoning. We introduce \textbf{UFO}, a benchmark designed to evaluate whether UFMs generate and use image and text cues as evidence for compositional multimodal reasoning. UFO spans three cue types, state determination, state reconstruction, and state augmentation, which correspond to progressively smaller transformations of the underlying world state. Our analysis reveals a significant modality gap, as models often achieve high prediction accuracy even when the generated visual cues exert limited influence on their decisions, indicating weakened evidential coupling and a reliance on textual shortcuts rather than robust cross modal grounding.
Applications · Energy
Open-source AI is scaling rapidly, and model hubs now host millions of artifacts. Each foundation model can spawn large numbers of fine-tunes, adapters, quantizations, merges, and forks. We take the position that compute efficiency alone is insufficient for sustainability in open-source AI. Lower per-run costs can accelerate experimentation and deployment, increasing aggregate footprint unless impacts are measurable and comparable across derivative lineages. However, the energy use, water consumption, and emissions of these derivative lineages are rarely measured or disclosed in a consistent, comparable way, leaving aggregate ecosystem impact largely invisible. We argue that sustainable open-source AI requires a coordination infrastructure that tracks impacts across model lineages, not only base models. We propose Data and Impact Accounting (DIA), a lightweight, non-restrictive transparency layer that (i) standardizes carbon-and-water reporting metadata, (ii) integrates low-friction measurement into common training and inference pipelines, and (iii) aggregates reports via public dashboards to summarize cumulative impacts across releases and derivatives. DIA makes derivative costs visible and supports ecosystem-level accountability while preserving openness.
Applications · Time Series
Existing machine learning models for network simulation excel at predicting average performance but, due to their reliance on mean squared error, systematically fail to capture the critical tail-latency and jitter that define modern network stability. This 'tail-blindness' renders them unreliable for latency-sensitive systems. We bridge this gap by introducing Accurate Neural Architecture Search (ANAS), a paradigm that automates the discovery of architectures for high-precision, distribution-aware network simulation. ANAS corrects the evaluation inaccuracies of weight-sharing NAS via a similarity-constrained search, employs a hybrid search space to model complex traffic, and uses a Wasserstein loss to optimize for the entire delay distribution, not just its mean. Empirically, the ANAS-discovered architecture is holistically superior: it reduces overall validation loss by 25.8\% compared to DeepQueueNet, demonstrating strong average-case performance, while simultaneously excelling at tail-sensitive metrics by lowering the normalized Wasserstein distance ($W_n$) by up to 69.8\%. This confirms its ability to faithfully model a comprehensive performance spectrum, encompassing both average and critical tail behaviors. The ANAS framework provides a practical methodology for automatically creating high-fidelity model of network devices, enabling more reliable validation of next-generation network protocols and algorithms.
Deep Learning · Foundation Models
This position paper argues that vision-language model (VLM) benchmarks for causal reasoning rely on two under-examined assumptions. First, benchmarks presuppose temporal constitution, the understanding of time as the medium through which causes produce effects, without testing it as a prerequisite. Second, they insufficiently distinguish external symbolic scaffolding from internalized capability; scaffolding-invariance is the diagnostic signature of genuine internalization. Drawing on frameworks from art, philosophy, and psychoanalysis, we propose diagnostics that probe these foundations. Preliminary evidence from three VLMs shows systematic disparity between fluent causal text and valid causal structure, and qualitatively different responses to identical scaffolding manipulation. None of these patterns indicates constitutive internalization. Progress requires benchmarks that test temporal understanding and scaffolding-invariance, not only output accuracy.