As large language models are increasingly utilized in real-world applications, guarantees of task-specific metrics are essential for their reliable deployment. Previous studies have introduced various criteria of conformal uncertainty grounded in split conformal prediction, which offer user-specified correctness coverage. However, existing frameworks often fail to identify uncertainty data outliers that violate the exchangeability assumption, leading to unbounded miscoverage rates and unactionable prediction sets. In this paper, we propose a novel approach termed Selective Conformal Uncertainty (SConU), which, for the first time, implements significance tests, by developing two conformal p-values that are instrumental in determining whether a given sample deviates from the uncertainty distribution of the calibration set at a specific manageable risk level. Our approach not only facilitates rigorous management of miscoverage rates across both single-domain and interdisciplinary contexts, but also enhances the efficiency of predictions. Furthermore, we comprehensively analyze the components of the conformal procedures, aiming to approximate conditional coverage, particularly in high-stakes question-answering tasks.
论文检索
输入标题、作者或关键词，从 1,669 篇学术成果中精准定位
Can’t See the Forest for the Trees: Benchmarking Multimodal Safety Awareness for Multimodal LLMs
PDF ↗Multimodal Large Language Models (MLLMs) have expanded the capabilities of traditional language models by enabling interaction through both text and images. However, ensuring the safety of these models remains a significant challenge, particularly in accurately identifying whether multimodal content is safe or unsafe—a capability we term safety awareness. In this paper, we introduce MMSafeAware, the first comprehensive multimodal safety awareness benchmark designed to evaluate MLLMs across 29 safety scenarios with 1,500 carefully curated image-prompt pairs. MMSafeAware includes both unsafe and over-safety subsets to assess models’ abilities to correctly identify unsafe content and avoid over-sensitivity that can hinder helpfulness. Evaluating nine widely used MLLMs using MMSafeAware reveals that current models are not sufficiently safe and often overly sensitive; for example, GPT-4V misclassifies 36.1% of unsafe inputs as safe and 59.9% of benign inputs as unsafe. We further explore three methods to improve safety awareness—prompting-based approaches, visual contrastive decoding, and vision-centric reasoning fine-tuning—but find that none achieve satisfactory performance. Our findings highlight the profound challenges in developing MLLMs with robust safety awareness, underscoring the need for further research in this area. All the code and data will be publicly available to facilitate future research.
Implicit hate speech has become a significant challenge for online platforms, as it often avoids detection by large language models (LLMs) due to its indirectly expressed hateful intent. This study identifies the limitations of LLMs in detecting implicit hate speech, particularly when disguised as seemingly harmless expressions in a rhetorical device. To address this challenge, we employ a Jailbreaking strategy and Energy-based Constrained Decoding techniques, and design a small model for measuring the energy of metaphorical rhetoric. This approach can lead to LLMs generating metaphorical implicit hate speech. Our research reveals that advanced LLMs, like GPT-4o, frequently misinterpret metaphorical implicit hate speech, and fail to prevent its propagation effectively. Even specialized models, like ShieldGemma and LlamaGuard, demonstrate inadequacies in blocking such content, often misclassifying it as harmless speech. This work points out the vulnerability of current LLMs to implicit hate speech, and emphasizes the improvements to address hate speech threats better.
Recent advancements in latent diffusion models (LDMs) have markedly enhanced text-to-audio generation, yet their iterative sampling processes impose substantial computational demands, limiting practical deployment. While recent methods utilizing consistency-based distillation aim to achieve few-step or single-step inference, their one-step performance is constrained by curved trajectories, preventing them from surpassing traditional diffusion models. In this work, we introduce FlashAudio with rectified flows to learn straight flow for fast simulation. To alleviate the inefficient timesteps allocation and suboptimal distribution of noise, FlashAudio optimizes the time distribution of rectified flow with Bifocal Samplers and proposes immiscible flow to minimize the total distance of data-noise pairs in a batch vias assignment. Furthermore, to address the amplified accumulation error caused by the classifier-free guidance (CFG), we propose Anchored Optimization, which refines the guidance scale by anchoring it to a reference trajectory. Experimental results on text-to-audio generation demonstrate that FlashAudio’s one-step generation performance surpasses the diffusion-based models with hundreds of sampling steps on audio quality and enables a sampling speed of 400x faster than real-time on a single NVIDIA 4090Ti GPU. Code will be available at https://github.com/liuhuadai/FlashAudio. Audio Samples are available at https://FlashAudio-TTA.github.io/.
Large language models (LLMs) exhibit the ability to generalize given few-shot examples in their input prompt, an emergent capability known as in-context learning (ICL). We investigate whether LLMs use ICL to perform structured reasoning in ways that are consistent with a Bayesian framework or rely on pattern matching. Using a controlled setting of biased coin flips, we find that: (1) LLMs often possess biased priors, causing initial divergence in zero-shot settings, (2) in-context evidence outweighs explicit bias instructions, (3) LLMs broadly follow Bayesian posterior updates, with deviations primarily due to miscalibrated priors rather than flawed updates, and (4) attention magnitude has negligible effect on Bayesian inference. With sufficient demonstrations of biased coin flips via ICL, LLMs update their priors in a Bayesian manner. Code and visualizations are available on the [project page](https://ai-climate.berkeley.edu/llm-coin-flips/).
People who frequently use ChatGPT for writing tasks are accurate and robust detectors of AI-generated text
PDF ↗In this paper, we study how well humans can detect text generated by commercial LLMs (GPT-4o, Claude, o1). We hire annotators to read 300 non-fiction English articles, label them as either human-written or AI-generated, and provide paragraph-length explanations for their decisions. Our experiments show that annotators who frequently use LLMs for writing tasks excel at detecting AI-generated text, even without any specialized training or feedback. In fact, the majority vote among five such “expert” annotators misclassifies only 1 of 300 articles, significantly outperforming most commercial and open-source detectors we evaluated even in the presence of evasion tactics like paraphrasing and humanization. Qualitative analysis of the experts’ free-form explanations shows that while they rely heavily on specific lexical clues (‘AI vocabulary’), they also pick up on more complex phenomena within the text (e.g., formality, originality, clarity) that are challenging to assess for automatic detectors. We release our annotated dataset and code to spur future research into both human and automated detection of AI-generated text.
Position · Social_ethical_env_impact
Patients have long sought health information online, and increasingly, they are turning to generative AI to answer their health-related queries. Given the high stakes of the medical domain, techniques like retrieval-augmented generation and citation grounding have been widely promoted as methods to reduce hallucinations and improve the accuracy of AI-generated responses and have been widely adopted into search engines. However, we argue that even when these methods produce literally accurate content drawn from source documents sans hallucinations, they can still be highly misleading. Patients may derive significantly different interpretations from AI-generated outputs than they would from reading the original source material, let alone consulting a knowledgeable clinician. Through a large-scale query analysis on topics including disputed diagnoses and procedure safety, we support our argument with quantitative and qualitative evidence of the suboptimal answers resulting from current systems. In particular, we highlight how these models tend to decontextualize facts, omit critical relevant sources, and reinforce patient misconceptions or biases. We propose a series of recommendations---such as the incorporation of communication pragmatics and enhanced comprehension of source documents---that could help mitigate these issues and extend beyond the medical domain.
Deep Learning · Large Language Models
Although large language models (LLMs) have shown promise in biomolecule optimization problems, they incur heavy computational costs and struggle to satisfy precise constraints. On the other hand, specialized solvers like LaMBO-2 offer efficiency and fine-grained control but require more domain expertise. Comparing these approaches is challenging due to expensive laboratory validation and inadequate synthetic benchmarks. We address this by introducing Ehrlich functions, a synthetic test suite that captures the geometric structure of biophysical sequence optimization problems. With prompting alone, off-the-shelf LLMs struggle to optimize Ehrlich functions. In response, we propose LLOME (Language Model Optimization with Margin Expectation), a bilevel optimization routine for online black-box optimization.When combined with a novel preference learning loss, we find LLOME can not only learn to solve some Ehrlich functions, but can even perform as well as or better than LaMBO-2 on moderately difficult Ehrlich variants. However, LLMs also exhibit some likelihood-reward miscalibration and struggle without explicit rewards. Our results indicate LLMs can occasionally provide significant benefits, but specialized solvers are still competitive and incur less overhead.
Theory · Learning Theory
This paper explores how theory can guide and enhance practical algorithms, using Low-Rank Adaptation (LoRA) (Hu et al., 2022) in large language models as a case study. We rigorously prove that, under gradient descent, LoRA adapters align with specific singular subspaces of the one-step full fine-tuning gradient. This result suggests that, by properly initializing the adapters using the one-step full gradient, subspace alignment can be achieved immediately—applicable to both linear and nonlinear models. Building on our theory, we propose a theory-driven algorithm, LoRA-One, where the linear convergence (as well as generalization) is built and incorporating preconditioners theoretically helps mitigate the effects of ill-conditioning. Besides, our theory reveals connections between LoRA-One and other gradient-alignment-based methods, helping to clarify misconceptions in the design of such algorithms. LoRA-One achieves significant empirical improvements over LoRA and its variants across benchmarks in natural language understanding, mathematical reasoning, and code generation. Code is available at: https://github.com/YuanheZ/LoRA-One.
Social Aspects · Robustness
As machine learning models are deployed in critical applications, robustness against adversarial perturbations is crucial. While numerous defensive algorithms have been proposed to counter such attacks, they typically assume that all adversarial transformations are equally important, an assumption that rarely aligns with real-world applications. To address this, we study the problem of robust learning against adversarial perturbations under cost-sensitive scenarios, where the potential harm of different types of misclassifications is encoded in a cost matrix. Our solution introduces a provably robust learning algorithm to certify and optimize for cost-sensitive robustness, building on the scalable certification framework of randomized smoothing. Specifically, we formalize the definition of cost-sensitive certified radius and propose our novel adaptation of the standard certification algorithm to generate tight robustness certificates tailored to any cost matrix. In addition, we design a robust training method that improves certified cost-sensitive robustness without compromising model accuracy. Extensive experiments on benchmark datasets, including challenging ones unsolvable by existing methods, demonstrate the effectiveness of our certification algorithm and training method across various cost-sensitive scenarios.
Theory · Learning Theory
This paper explores how theory can guide and enhance practical algorithms, using Low-Rank Adaptation (LoRA) (Hu et al., 2022) in large language models as a case study. We rigorously prove that, under gradient descent, LoRA adapters align with specific singular subspaces of the one-step full fine-tuning gradient. This result suggests that, by properly initializing the adapters using the one-step full gradient, subspace alignment can be achieved immediately—applicable to both linear and nonlinear models. Building on our theory, we propose a theory-driven algorithm, LoRA-One, where the linear convergence (as well as generalization) is built and incorporating preconditioners theoretically helps mitigate the effects of ill-conditioning. Besides, our theory reveals connections between LoRA-One and other gradient-alignment-based methods, helping to clarify misconceptions in the design of such algorithms. LoRA-One achieves significant empirical improvements over LoRA and its variants across benchmarks in natural language understanding, mathematical reasoning, and code generation. Code is available at: https://github.com/YuanheZ/LoRA-One.
Applications · Everything Else
Implicit feedback recommendation is challenged by the missing negative feedback essential for effective model training. Existing methods often resort to negative sampling, a technique that assumes unlabeled interactions as negative samples. This assumption risks misclassifying potential positive samples within the unlabeled data, thereby undermining model performance. To address this issue, we introduce PURL, a model-agnostic framework that reframes implicit feedback recommendation as a weakly supervised learning task, eliminating the need for negative samples. However, its unbiasedness hinges on the accurate estimation of the class prior. To address this challenge, we propose Progressive Proximal Transport (PPT), which estimates the class prior by minimizing the proximal transport cost between positive and unlabeled samples. Experiments on three real-world datasets validate the efficacy of PURL in terms of improved recommendation quality. Code is available at https://github.com/HowardZJU/weakrec.
Theory · Optimization
Binary decision diagrams (BDDs) are widely applied tools to compactly represent labeled data as directed acyclic graphs; for efficiency and interpretability reasons small BDDs are preferred.Given labeled data, minimizing BDDs is NP-complete and thus recent research focused on the influence of parameters such as the solution size $s$ on the complexity [Ordyniak et al., AAAI 2024].Our main positive result is an algorithm that is efficient if in particular $s$, the domain size $D$, and the Hamming distance between any two data points is small, improving on previous running-time bounds.This algorithm is inspired by the witness-tree paradigm that was recently successful for computing decision trees [Komusiewicz et al., ICML 2023], whose extension to BDDs was open.We extend our algorithmic results to the case where we allow a small number of misclassified data points and complement them with lower bounds that show that the running times are tight from multiple points of view.We show that our main algorithm holds practical promise by providing a proof-of-concept implementation.
General Machine Learning · Representation Learning
Machine learning models often perform well on tabular data by optimizing average prediction accuracy. However, they may underperform on specific subsets due to inherent biases and spurious correlations in the training data, such as associations with non-causal features like demographic information. These biases lead to critical robustness issues as models may inherit or amplify them, resulting in poor performance where such misleading correlations do not hold. Existing mitigation methods have significant limitations: some require prior group labels, which are often unavailable, while others focus solely on the conditional distribution \(P(Y|X)\), upweighting misclassified samples without effectively balancing the overall data distribution \(P(X)\). To address these shortcomings, we propose a latent score-based reweighting framework. It leverages score-based models to capture the joint data distribution \(P(X, Y)\) without relying on additional prior information. By estimating sample density through the similarity of score vectors with neighboring data points, our method identifies underrepresented regions and upweights samples accordingly. This approach directly tackles inherent data imbalances, enhancing robustness by ensuring a more uniform dataset representation. Experiments on various tabular datasets under distribution shifts demonstrate that our method effectively improves performance on imbalanced data.
Deep Learning · Large Language Models
Multimodal large language models (MLLMs) have achieved impressive progress in tasks such as visual question answering and visual understanding, but they still face significant challenges in emotional reasoning. Current methods to enhance emotional understanding typically rely on fine-tuning or manual annotations, which are resource-intensive and limit scalability. In this work, we focus on improving the ability of MLLMs to capture emotions during the inference phase. Specifically, MLLMs encounter two main issues: they struggle to distinguish between semantically similar emotions, leading to misclassification, and they are overwhelmed by redundant or irrelevant visual information, which distracts from key emotional cues. To address these, we propose Sharpening Emotion Perception in MLLMs (SEPM), which incorporates a Confidence-Guided Coarse-to-Fine Inference framework to refine emotion classification by guiding the model through simpler tasks. Additionally, SEPM employs Focus-on-Emotion Visual Augmentation to reduce visual redundancy by directing the attention of models to relevant emotional cues in images. Experimental results demonstrate that SEPM significantly improves MLLM performance on emotion-related tasks, providing a resource-efficient and scalable solution for emotion recognition.
Theory · Game Theory
Strategic classification addresses a learning problem where a decision-maker implements a classifier over agents who may manipulate their features in order to receive favorable predictions. In the standard model of online strategic classification, in each round, the decision-maker implements and publicly reveals a classifier, after which agents perfectly best respond based on this knowledge. However, in practice, whether to disclose the classifier is often debated---some decision-makers believe that hiding the classifier can prevent misclassification errors caused by manipulation. In this paper, we formally examine how limiting the agents' access to the current classifier affects the decision-maker's performance. Specifically, we consider an extended online strategic classification setting where agents lack direct knowledge about the current classifier and instead manipulate based on a weighted average of historically implemented classifiers. Our main result shows that in this setting, the decision-maker incurs $(1-\gamma)^{-1}$ or $k_{\text{in}}$ times more mistakes compared to the full-knowledge setting, where $k_{\text{in}}$ is the maximum in-degree of the manipulation graph (representing how many distinct feature vectors can be manipulated to appear as a single one), and $\gamma$ is the discount factor indicating agents' memory of past classifiers. Our results demonstrate how withholding access to the classifier can backfire and degrade the decision-maker's performance in online strategic classification.
Social Aspects · Safety
Jailbreaking techniques trick Large Language Models (LLMs) into producing restricted output, posing a potential threat. One line of defense is to use another LLM as a Judge to evaluate the harmfulness of generated text. However, we reveal that these Judge LLMs are vulnerable to token segmentation bias, an issue that arises when delimiters alter the tokenization process, splitting words into smaller sub-tokens. This alters the embeddings of the entire sequence, reducing detection accuracy and allowing harmful content to be misclassified as safe. In this paper, we introduce Emoji Attack, a novel strategy that amplifies existing jailbreak prompts by exploiting token segmentation bias. Our method leverages in-context learning to systematically insert emojis into text before it is evaluated by a Judge LLM, inducing embedding distortions that significantly lower the likelihood of detecting unsafe content. Unlike traditional delimiters, emojis also introduce semantic ambiguity, making them particularly effective in this attack. Through experiments on state-of-the-art Judge LLMs, we demonstrate that Emoji Attack substantially reduces the unsafe prediction rate, bypassing existing safeguards.
General Machine Learning · Clustering
In this paper, we investigate the problem of recovering hidden communities in the Labeled Stochastic Block Model (LSBM) with a finite number of clusters whose sizes grow linearly with the total number of nodes. We derive the necessary and sufficient conditions under which the expected number of misclassified nodes is less than $ s $, for any number $ s = o(n) $. To achieve this, we propose IAC (Instance-Adaptive Clustering), the first algorithm whose performance matches the instance-specific lower bounds both in expectation and with high probability.IAC is a novel two-phase algorithm that consists of a one-shot spectral clustering step followed by iterative likelihood-based cluster assignment improvements. This approach is based on the instance-specific lower bound and notably does not require any knowledge of the model parameters, including the number of clusters. By performing the spectral clustering only once, IAC maintains an overall computational complexity of $ \mathcal{O}(n\, \text{polylog}(n)) $, making it scalable and practical for large-scale problems.
Deep Learning · Sequential Models, Time series
Spiking Neural Networks (SNNs) are considered promising energy-efficient models due to their dynamic capability to process spatial-temporal spike information. Existing work has demonstrated that SNNs exhibit temporal heterogeneity, which leads to diverse outputs of SNNs at different time steps and has the potential to enhance their performance. Although SNNs obtained by direct training methods achieve state-of-the-art performance, current methods introduce limited temporal heterogeneity through the dynamics of spiking neurons or network structures. They lack the improvement of temporal heterogeneity through the lens of the gradient. In this paper, we first conclude that the diversity of the temporal logit gradients in current methods is limited. This leads to insufficient temporal heterogeneity and results in temporally miscalibrated SNNs with degraded performance. Based on the above analysis, we propose a Temporal Model Calibration (TMC) method, which can be seen as a logit gradient rescaling mechanism across time steps. Experimental results show that our method can improve the temporal logit gradient diversity and generate temporally calibrated SNNs with enhanced performance. In particular, our method achieves state-of-the-art accuracy on ImageNet, DVSCIFAR10, and N-Caltech101. Codes are available at https://github.com/zju-bmi-lab/TMC.
Position · Methodology
To detect distribution shifts and improve model safety, many out-of-distribution (OOD) detection methods rely on the predictive uncertainty or features of supervised models trained on in-distribution data. In this position paper, we critically re-examine this popular family of OOD detection procedures, and we argue that these methods are fundamentally answering the wrong questions for OOD detection. There is no simple fix to this misalignment, since a classifier trained only on in-distribution classes cannot be expected to identify OOD points; for instance, a cat-dog classifier may confidently misclassify an airplane if it contains features that distinguish cats from dogs, despite generally appearing nothing alike. We find that uncertainty-based methods incorrectly conflate high uncertainty with being OOD, while feature-based methods incorrectly conflate far feature-space distance with being OOD. We show how these pathologies manifest as irreducible errors in OOD detection and identify common settings where these methods are ineffective. Additionally, interventions to improve OOD detection such as feature-logit hybrid methods, scaling of model and data size, epistemic uncertainty representation, and outlier exposure also fail to address this fundamental misalignment in objectives.