Event cameras are renowned for their high efficiency due to outputting a sparse, asynchronous stream of events. However, they are plagued by noisy events, especially in low light conditions. Denoising is an essential task for event cameras, but evaluating denoising performance is challenging. Label-dependent denoising metrics involve artificially adding noise to clean sequences, complicating evaluations. Moreover, the majority of these metrics are monotonic, which can inflate scores by removing substantial noise and valid events. To overcome these limitations, we propose the first label-free and non-monotonic evaluation metric, the area of the continuous contrast curve (AOCC), which utilizes the area enclosed by event frame contrast curves across different time intervals. This metric is inspired by how events capture the edge contours of scenes or objects with high temporal resolution. An effective denoising method removes noise without eliminating these edge-contour events, thus preserving the contrast of event frames. Consequently, contrast across various time ranges serves as a metric to assess denoising effectiveness. As the time interval lengthens, the curve will initially rise and then fall. The proposed metric is validated through both theoretical and experimental evidence. The codes are available at https://github.com/shicy17/AOCC.

Many stakeholders---from model developers to policymakers---seek to minimize the risks of large language models (LLMs). Key to this goal is whether technical safeguards can impede the misuse of LLMs, even when models are customizable via fine-tuning or when model weights are openly available. Several recent studies have proposed methods to produce durable LLM safeguards for open-weight LLMs that can withstand adversarial modifications of the model's weights via fine-tuning. This holds the promise of raising adversaries' costs even under strong threat models where adversaries can directly fine-tune parameters. However, we caution against over-reliance on such methods in their current state. Through several case studies, we demonstrate that even the evaluation of these defenses is exceedingly difficult and can easily mislead audiences into thinking that safeguards are more durable than they really are. We draw lessons from the failure modes that we identify and suggest that future research carefully cabin claims to more constrained, well-defined, and rigorously examined threat models, which can provide useful and candid assessments to stakeholders.

Evaluating aligned large language models' (LLMs) ability to recognize and reject unsafe user requests is crucial for safe, policy-compliant deployments. Existing evaluation efforts, however, face three limitations that we address with **SORRY-Bench**, our proposed benchmark. **First**, existing methods often use coarse-grained taxonomies of unsafe topics, and are over-representing some fine-grained topics. For example, among the ten existing datasets that we evaluated, tests for refusals of self-harm instructions are over 3x less represented than tests for fraudulent activities. SORRY-Bench improves on this by using a fine-grained taxonomy of 44 potentially unsafe topics, and 440 class-balanced unsafe instructions, compiled through human-in-the-loop methods. **Second**, evaluations often overlook the linguistic formatting of prompts, like different languages, dialects, and more --- which are only implicitly considered in many evaluations. We supplement SORRY-bench with 20 diverse linguistic augmentations to systematically examine these effects. **Third**, existing evaluations rely on large LLMs (e.g., GPT-4) for evaluation, which can be computationally expensive. We investigate design choices for creating a fast, accurate automated safety evaluator. By collecting 7K+ human annotations and conducting a meta-evaluation of diverse LLM-as-a-judge designs, we show that fine-tuned 7B LLMs can achieve accuracy comparable to GPT-4 scale LLMs, with lower computational cost. Putting these together, we evaluate over 50 proprietary and open-weight LLMs on SORRY-Bench, analyzing their distinctive safety refusal behaviors. We hope our effort provides a building block for systematic evaluations of LLMs' safety refusal capabilities, in a balanced, granular, and efficient manner. Benchmark demo, data, code, and models are available through [https://sorry-bench.github.io](https://sorry-bench.github.io).

Detecting transient radiant expanding flares is crucial for assessing the status of key devices like Hall thrusters, for which real-time detection is vital to maintain satellite stability in orbit. Event cameras are better suited for this task than high-speed cameras, offering faster perception speed, lower power consumption, and more compact sizes. While previous event-based works have significantly enhanced the visual processing speed, current approaches still fail to meet real-time requirements for transient flare detection, due to complex algorithms and inefficient hardware designs. To address the issues, we propose ROFD, an event-based 5k-fps real-time optical flow detector that includes: 1) A residual spatio-temporal-average optical flow detection algorithm that reduces the computing complexity and shortens the detection time window. 2) A tile-based interleaving memory mapping method that minimizes wasted memory access time. 3) Conflict-free data flows that eliminate data dependency and enhance parallelism. Experiments demonstrate that our FPGA-implemented ROFD operates at 5k-fps, achieving a maximum speedup of 35.4 × and a maximum accuracy improvement of 2.98×, while saving 87% DSPs, 33% BRAMs, and 58% power consumption compared to SOTA.

Foundation models are increasingly becoming better autonomous programmers, raising the prospect that they could also automate dangerous offensive cyber‑operations. Current frontier model audits probe the cybersecurity risks of such agents, but most fail to account for the degrees of freedom available to adversaries in the real world.In particular, with strong verifiers and financial incentives, agents for offensive cybersecurity are amenable to iterative improvement by would-be adversaries. We argue that assessments should take into account an expanded threat model in the context of cybersecurity, emphasizing the varying degrees of freedom that an adversary may possess in stateful and non-stateful environments within a fixed compute budget. We show that even with a relatively small compute budget (8 H100 GPU Hours in our study), adversaries can improve an agent's cybersecurity capability on InterCode CTF by more than 40\% relative to the baseline---without any external assistance. These results highlight the need to evaluate agents' cybersecurity risk in a dynamic manner, painting a more representative picture of risk.

Event cameras, which are highly sensitive to light intensity changes, often generate substantial noise during imaging. Existing denoising methods either lack the speed for real-time processing or struggle with dynamic scenes, mistakenly discarding valid events. To address these issues, we propose a novel dual-stage polarity-focused denoising (PFD) method that leverages the consistency of polarity and its changes within local pixel areas. Whether due to camera motion or dynamic scene changes, the polarity and its changes in triggered events are highly correlated with these movements, allowing for effective noise handling. We introduce two versions: PFD-A, which excels at reducing background activity (BA) noise, and PFD-B, which is designed to address both BA and flicker noise. Both versions are lightweight and computationally efficient. The experimental results show that PFD outperforms benchmark methods in terms of the SNR and ESR metrics, achieving state-of-the-art performance across various datasets. Additionally, we propose an FPGA implementation of PFD processes that handles each event in just 7 clock cycles, ensuring real-time performance. The codes are available athttps://github.com/shicy17/PFD.

Large language models (LLMs) show inherent brittleness in their safety mechanisms, as evidenced by their susceptibility to jailbreaking and even non-malicious fine-tuning. This study explores this brittleness of safety alignment by leveraging pruning and low-rank modifications. We develop methods to identify critical regions that are vital for safety guardrails, and that are disentangled from utility-relevant regions at both the neuron and rank levels. Surprisingly, the isolated regions we find are sparse, comprising about $3$ % at the parameter level and $2.5$ % at the rank level. Removing these regions compromises safety without significantly impacting utility, corroborating the inherent brittleness of the model's safety mechanisms. Moreover, we show that LLMs remain vulnerable to low-cost fine-tuning attacks even when modifications to the safety-critical regions are restricted. These findings underscore the urgent need for more robust safety strategies in LLMs.

Language models (LMs) derive their capabilities from extensive training on diverse data, including copyrighted material. These models can memorize and generate content similar to their training data, potentially risking legal issues like copyright infringement.Therefore, model creators are motivated to develop mitigation methods that prevent generating particular copyrighted content, an ability we refer to as copyright takedowns. This paper introduces the first evaluation of the feasibility and side effects of copyright takedowns for LMs. We propose CoTaEval, an evaluation framework to assess the effectiveness of copyright takedown methods,the impact on the model's ability to retain uncopyrightable factual knowledge from the copyrighted content, and how well the model maintains its general utility and efficiency.We examine several strategies, including adding system prompts, decoding-time filtering interventions, and unlearning approaches. Our findings indicate that no method excels across all metrics, showing significant room for research in this unique problem setting and indicating potential unresolved challenges for live policy proposals.

Light interference negatively impacts on frame-based visual tasks. Phenomena such as overexposure cause the loss of valuable information and reduce task execution efficiency. Event cameras are neuromorphic vision sensors that output sparse, asynchronous streams of events rather than frames. These cameras feature high temporal resolution, high dynamic range, and low power consumption. As a result, they are not susceptible to overexposure and motion blur, and they are able to recognize light interference such as strobe lights, stray lights, and reflections. However, event cameras are highly sensitive to light intensity changes, so light interference still affects event cameras as noise which easily alias with events triggered by environmental objects. Therefore, to reduce or eliminate the negative impact of light interference on event cameras, we systematically analyze the optical properties and event-triggering principles of these forms of light interference, and then propose ELIR (Event-based Light Interference Removal) method for removing light interference signals in event streams under static and dynamic scenes. The proposed method is validated in object detection tasks. Additionally, we launch the LIED datasets to evaluate the effect of light interference removal in event streams to assist with other studies in this field. Experimental results on the LIED datasets show that our proposed method can remove, on average, over 97% of light interference in static scenes, over 86% in dynamic scenes. Finally, the proposed method is verified on the object detection task, achieving an average PRE over 92%. The dataset is available at https://github.com/shicy17/LIED.