Kai Zou, Ziqi Huang, Yuhao Dong, Shulin Tian, Dian Zheng, Hongbo Liu, Jingwen He, Bin Liu, Yu Qiao, Ziwei Liu. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Recent advancements in visual generative models have enabled high-quality image and video generation, opening diverse applications. However, evaluating these models often demands sampling hundreds or thousands of images or videos, making the process computationally expensive, especially for diffusion-based models with inherently slow sampling. Moreover, existing evaluation methods rely on rigid pipelines that overlook specific user needs and provide numerical results without clear explanations. In contrast, humans can quickly form impressions of a model’s capabilities by observing only a few samples. To mimic this, we propose the Evaluation Agent framework, which employs human-like strategies for efficient, dynamic, multi-round evaluations using only a few samples per round, while offering detailed, user-tailored analyses. It offers four key advantages: 1) efficiency, 2) promptable evaluation tailored to diverse user needs, 3) explainability beyond single numerical scores, and 4) scalability across various models and tools. Experiments show that Evaluation Agent reduces evaluation time to 10% of traditional methods while delivering comparable results. The Evaluation Agent framework is fully open-sourced to advance research in visual generative models and their efficient evaluation.

Robotic manipulation in open-world settings requires not only task execution but also the ability to detect and learn from failures. While recent advances in vision-language models (VLMs) and large language models (LLMs) have improved robots' spatial reasoning and problem-solving abilities, they still struggle with failure recognition, limiting their real-world applicability. We introduce AHA, an open-source VLM designed to detect and reason about failures in robotic manipulation using natural language. By framing failure detection as a free-form reasoning task, AHA identifies failures and provides detailed, adaptable explanations across different robots, tasks, and environments. We fine-tuned AHA using FailGen, a scalable framework that generates the first large-scale dataset of robotic failure trajectories, the AHA dataset. FailGen achieves this by procedurally perturbing successful demonstrations from simulation. Despite being trained solely on the AHA dataset, AHA generalizes effectively to real-world failure datasets, robotic systems, and unseen tasks. It surpasses the second-best model (GPT-4o in-context learning) by 10.3% and exceeds the average performance of six compared models including five state-of-the-art VLMs by 35.3% across multiple metrics and datasets. We integrate AHA into three manipulation frameworks that utilize LLMs/VLMs for reinforcement learning, task and motion planning, and zero-shot trajectory generation. AHA’s failure feedback enhances these policies' performances by refining dense reward functions, optimizing task planning, and improving sub-task verification, boosting task success rates by an average of 21.4% across all three tasks compared to GPT-4 models. Project page: https://aha-vlm.github.io

Low-light image enhancement task is essential yet challenging as it is ill-posed intrinsically. Previous arts mainly focus on the low-light images captured in the visible spectrum using pixel-wise loss, which limits the capacity of recovering the brightness, contrast, and texture details due to the small number of income photons. In this work, we propose a novel approach to increase the visibility of images captured under low-light environments by removing the in-camera infrared (IR) cut-off filter, which allows for the capture of more photons and results in improved signal-to-noise ratio due to the inclusion of information from the IR spectrum. To verify the proposed strategy, we collect a paired dataset of low-light images captured without the IR cut-off filter, with corresponding long-exposure reference images with an external filter. The experimental results on the proposed dataset demonstrate the effectiveness of the proposed method, showing better performance quantitatively and qualitatively. The dataset and code are publicly available at https://wyf0912.github.io/ELIEI/

Abstract The time‐interleaved analog‐to‐digital converter (TIADC) system could achieve high sampling rate effectively, but the frequency response inconsistency among channels will introduce frequency response mismatches, which will seriously degrade the signal quality. In this paper, a novel multi‐channel frequency response mismatches adaptive calibration method is proposed. The frequency response of the compensation filter is expanded into a polynomial of a digital differentiator, and the least mean square adaptive strategy is used to calibrate frequency response mismatches. It does not require measuring the channel frequency responses in advance and uses digital post‐processing technology to effectively compensate the frequency response of the multi‐channel TIADC system and improve the spurious‐free dynamic range. The experimental results verify the effectiveness of the algorithm.

Abstract A maximum correntropy criterion‐blind clustering multi‐modulus algorithm (MCC‐BCMMA) is proposed to equalize multi‐level quadrature amplitude modulation (QAM) channels in impulsive noise environment. The novel cost function is based on the maximum correntropy criterion (MCC), theoretical analysis shows that this criterion is useful when the noise is impulsive. Moreover, this cost function is decomposed into in‐phase term and orthogonal term to eliminate the phase ambiguity, treats high‐order QAM signal to classical QAM4 signal to reduce the error, and calculates the weight coefficients by a simple way. Simulation results show that the newly proposed algorithm MCC‐BCMMA can be used to equalize multi‐level QAM signals in both Gaussian and impulsive noise environments, and it shows better equalization performance under impulsive noise. The relationship between the kernel size of MCC, the convergence speed and accuracy is also analysed.

Jitter is one of the key factors affecting bit error rate (BER) on high-speed links. A novel method of jitter decomposition by PointNet using 2D point cloud of jitter histogram is proposed for decomposing the time interval error (TIE) jitter into deterministic jitter (DJ) and random jitter (RJ). The proposed method uses transformNet (T-Net) and multi-layer perceptron (MLP) to learn global point cloud features and uses average pooling to aggregate information from all the points. Experimental results show that the proposed approach can decompose jitter into DJ and RJ, and is better than the traditional jitter decomposition method i.e. time lag correlation (TLC) and Time-Domain PLL Jitter Decomposition (T-D PLL). In addition, results show that the performance of the proposed method is better than CNN, PointRNN, and PointANN.

Abstract In this paper, a novel source signal recovery method for fast time‐varying channels described by complex exponential‐basis expansion model (CE‐BEM) in the impulsive noise environment is proposed. This method consists of two phases. The first phase is the equalization of fast time‐varying channels, in this phase, a novel algorithm FSE‐FLOS‐CMA is proposed. The convergence performance of this newly proposed algorithm is much better than that of existing fractionally spaced equalizer‐constant modulus algorithm (FSE‐CMA) in impulsive noise environment. In the second phase, a novel frequency estimation method is proposed. The estimated frequency in impulsive noise environment is obtained by calculating a specific p‐order fractional low‐order cyclic moment of the equalized signal. Simulation results show that the proposed FSE‐FLOS‐CMA and frequency estimation method can effectively estimate the source signals transmitted through the fast time‐varying channels in impulsive noise environment.

Cardiovascular diseases (CVDs) have become a serious hidden threat to human health. Because they are chronic circulatory diseases, we should detect and prevent them in real time to minimize damage to human health. The ideal solution to overcome the problems of the clinical biochemical analyzer is to connect the biochemical sensor to the Internet. The broadcast television network has access to the broadband network through the set-top box. At present, the market is undergoing rapid development. Furthermore, with the introduction of Industry 4.0, cyber-physical systems have become key framework technology, as their potential for implementation in smart home applications is high. Through the use of a typical household universal television device, we linked a total cholesterol and triglyceride test strip with a television controller to realize a method for remote cardiovascular disease management and prevention. Additionally, a cyber-physical healthcare system was added to the broadcast television network for in-home CVD diagnosis and family health management. By exploiting the portable, fast, and simple features afforded by the Internet of Things, the proposed remote medical device has satisfied the family monitoring requirements, and is expected to be widely applied in the near future.

In attempting to overcome the limitation of current methods to solve complicated constrained optimization problems, this paper proposes an adaptive hybrid particle swarm optimization multi-objective optimization (AHPSOMO) algorithm. In the early stage, this algorithm initializes the individuals in a population in an even manner using good point set (GPS) theory so that the diversity of the population can be guaranteed. In the process of local search, differential evolution (DE) algorithm is introduced for updating local optimal individuals. Particle swarm optimization method is further adopted to conduct global search as per the multi-objective approach. The results of simulation tests on 24 classic test functions and three engineering constrained optimization problems show that compared with other algorithms, our proposed algorithm is effective and feasible, which can offer highly accurate solutions with good robustness.

The paper proposes a fault diagnosis model based on the HIWO–SVM algorithm given the fact that the basic support vector machines (SVM) cannot solve effectively the problem of fault diagnosis in analog circuit. First of all, the wavelet package technique is adopted for extracting the information of the faults from the test points in the analog circuit. The differential evolution (DE) algorithm is then integrated with the purpose of improving the performance of the basic IWO algorithm, i.e. a hybrid IWO (HIWO) algorithm. The HIWO algorithm is further used to optimize the parameters of SVM in order to avoid the randomness of the parameter selection, thereby improving the diagnosis precision and robustness. The experimental results on a filter circuit show that the method is more effective and reliable than the other methods for fault diagnosis.