DiscoVLA: Discrepancy Reduction in Vision, Language, and Alignment for Parameter-Efficient Video-Text Retrieval

The parameter-efficient adaptation of the image-text pretraining model CLIP for video-text retrieval is a prominent area of research. While CLIP is focused on image-level vision-language matching, video-text retrieval demands comprehensive understanding at the video level. Three key discrepancies emerge in the transfer from image-level to video-level: vision, language, and alignment. However, existing methods mainly focus on vision while neglecting language and alignment. In this paper, we propose Discrepancy Reduction in Vision, Language, and Alignment (DiscoVLA), which simultaneously mitigates all three discrepancies. Specifically, we introduce Image-Video Features Fusion to integrate image-level and video-level features, effectively tackling both vision and language discrepancies. Additionally, we generate pseudo image captions to learn fine-grained image-level alignment. To mitigate alignment discrepancies, we propose Image-to-Video Alignment Distillation, which leverages image-level alignment knowledge to enhance video-level alignment. Extensive experiments demonstrate the superiority of our DiscoVLA. In particular, on MSRVTT with CLIP (ViT-B/16), DiscoVLA outperforms previous methods by 1.5% in R@1, reaching a final score of 50.5% R@1. The code is available at https://github.com/LunarShen/DsicoVLA.

Preference-Driven Multi-Objective Combinatorial Optimization with Conditional Computation

Recent deep reinforcement learning methods have achieved remarkable success in solving multi-objective combinatorial optimization problems (MOCOPs) by decomposing them into multiple subproblems, each associated with a specific weight vector. However, these methods typically treat all subproblems equally and solve them using a single model, hindering the effective exploration of the solution space and thus leading to suboptimal performance. To overcome the limitation, we propose POCCO, a novel plug-and-play framework that enables adaptive selection of model structures for subproblems, which are subsequently optimized based on preference signals rather than explicit reward values. Specifically, we design a conditional computation block that routes subproblems to specialized neural architectures. Moreover, we propose a preference-driven optimization algorithm that learns pairwise preferences between winning and losing solutions. We evaluate the efficacy and versatility of POCCO by applying it to two state-of-the-art neural methods for MOCOPs. Experimental results across four classic MOCOP benchmarks demonstrate its significant superiority and strong generalization.

FastVID: Dynamic Density Pruning for Fast Video Large Language Models

Video Large Language Models have shown impressive capabilities in video comprehension, yet their practical deployment is hindered by substantial inference costs caused by redundant video tokens. Existing pruning techniques fail to fully exploit the spatiotemporal redundancy inherent in video data. To bridge this gap, we perform a systematic analysis of video redundancy from two perspectives: temporal context and visual context. Leveraging this insight, we propose Dynamic Density Pruning for Fast Video LLMs termed FastVID. Specifically, FastVID dynamically partitions videos into temporally ordered segments to preserve temporal structure and applies a density-based token pruning strategy to maintain essential visual information. Our method significantly reduces computational overhead while maintaining temporal and visual integrity. Extensive evaluations show that FastVID achieves state-of-the-art performance across various short- and long-video benchmarks on leading Video LLMs, including LLaVA-OneVision and LLaVA-Video. Notably, FastVID effectively prunes 90% of video tokens while retaining 98.0% of LLaVA-OneVision's original performance. The code is available at https://github.com/LunarShen/FastVID.

Unicorn: A Universal and Collaborative Reinforcement Learning Approach Towards Generalizable Network-Wide Traffic Signal Control

Adaptive traffic signal control (ATSC) is crucial in reducing congestion, maximizing throughput, and improving mobility in rapidly growing urban areas. Recent advancements in parameter-sharing multi-agent reinforcement learning (MARL) have greatly enhanced the scalable and adaptive optimization of complex, dynamic flows in large-scale homogeneous networks. However, the inherent heterogeneity of real-world traffic networks, with their varied intersection topologies and interaction dynamics, poses substantial challenges to achieving scalable and effective ATSC across different traffic scenarios. To address these challenges, we present Unicorn, a universal and collaborative MARL framework designed for efficient and adaptable network-wide ATSC. Specifically, we first propose a unified approach to map the states and actions of intersections with varying topologies into a common structure based on traffic movements. Next, we design a Universal Traffic Representation (UTR) module with a decoder-only network for general feature extraction, enhancing the model's adaptability to diverse traffic scenarios. Additionally, we incorporate an Intersection Specifics Representation (ISR) module, designed to identify key latent vectors that represent the unique intersection's topology and traffic dynamics through variational inference techniques. To further refine these latent representations, we employ a contrastive learning approach in a self-supervised manner, which enables better differentiation of intersection-specific features. Moreover, we integrate the state-action dependencies of neighboring agents into policy optimization, which effectively captures dynamic agent interactions and facilitates efficient regional collaboration. Our results show that Unicorn outperforms other methods across various evaluation metrics, highlighting its potential in complex, dynamic traffic networks.

LLaVA-MLB: Mitigating and Leveraging Attention Bias for Training-Free Video LLMs

Training-free video large language models (LLMs) leverage pretrained Image LLMs to process video content without the need for further training. A key challenge in such approaches is the difficulty of retaining essential visual and temporal information, constrained by the token limits in Image LLMs. To address this, we propose a two-stage method for selecting query-relevant tokens based on the LLM attention scores: compressing the video sequence and then expanding the sequence. However, during the compression stage, Image LLMs often exhibit a positional attention bias in video sequences, where attention is overly concentrated on later frames, causing early-frame information to be underutilized. To alleviate this attention bias during sequence compression, we propose Gridded Attention Pooling for preserving spatiotemporal structure. Additionally, we introduce Visual Summarization Tail to effectively utilize this bias, facilitating overall video understanding during sequence expansion. In this way, our method effectively Mitigates and Leverages attention Bias (LLaVA-MLB), enabling the frozen Image LLM for detailed video understanding. Experiments on several benchmarks demonstrate that our approach outperforms state-of-the-art methods, achieving superior performance in both efficiency and accuracy. Our code will be released.

Detecting Defective Wafers Via Modular Networks

The growing availability of sensors within semiconductor manufacturing processes makes it feasible to detect defective wafers with data-driven models. Without directly measuring the quality of semiconductor devices, they capture the modalities between diverse sensor readings and can be used to predict key quality indicators (KQI, \textit{e.g.}, roughness, resistance) to detect faulty products, significantly reducing the capital and human cost in maintaining physical metrology steps. Nevertheless, existing models pay little attention to the correlations among different processes for diverse wafer products and commonly struggle with generalizability issues. To enable generic fault detection, in this work, we propose a modular network (MN) trained using time series stage-wise datasets that embodies the structure of the manufacturing process. It decomposes KQI prediction as a combination of stage modules to simulate compositional semiconductor manufacturing, universally enhancing faulty wafer detection among different wafer types and manufacturing processes. Extensive experiments demonstrate the usefulness of our approach, and shed light on how the compositional design provides an interpretable interface for more practical applications.

TempMe: Video Temporal Token Merging for Efficient Text-Video Retrieval

Most text-video retrieval methods utilize the text-image pre-trained CLIP as a backbone, incorporating complex modules that result in high computational overhead. As a result, many studies focus on efficient fine-tuning. The primary challenge in efficient adaption arises from the inherent differences between image and video modalities. Each sampled video frame must be processed by the image encoder independently, which increases complexity and complicates practical deployment. Although existing efficient methods fine-tune with small trainable parameters, they still incur high inference costs due to the large token number. In this work, we argue that temporal redundancy significantly contributes to the model's high complexity due to the repeated information in consecutive frames. Existing token compression methods for image models fail to solve the unique challenges, as they overlook temporal redundancy across frames. To tackle these problems, we propose Temporal Token Merging (TempMe) to reduce temporal redundancy. Specifically, we introduce a progressive multi-granularity framework. By gradually combining neighboring clips, we merge temporal tokens across different frames and learn video-level features, leading to lower complexity and better performance. Extensive experiments validate the superiority of our TempMe. Compared to previous efficient text-video retrieval methods, TempMe significantly reduces output tokens by 95% and GFLOPs by 51%, while achieving a 1.8X speedup and a 4.4% R-Sum improvement. Additionally, TempMe exhibits robust generalization capabilities by integrating effectively with both efficient and full fine-tuning methods. With full fine-tuning, TempMe achieves a significant 7.9% R-Sum improvement, trains 1.57X faster, and utilizes 75.2% GPU memory usage. Our code will be released.

SSD4Rec: A Structured State Space Duality Model for Efficient Sequential Recommendation

Sequential recommendation methods are crucial in modern recommender systems for their remarkable capability to understand a user's changing interests based on past interactions. However, a significant challenge faced by current methods (e.g., RNN- or Transformer-based models) is to effectively and efficiently capture users' preferences by modeling long behavior sequences, which impedes their various applications like short video platforms where user interactions are numerous. Recently, an emerging architecture named Mamba, built on state space models (SSM) with efficient hardware-aware designs, has showcased the tremendous potential for sequence modeling, presenting a compelling avenue for addressing the challenge effectively. Inspired by this, we propose a novel generic and efficient sequential recommendation backbone, SSD4Rec, which explores the seamless adaptation of Mamba for sequential recommendations. Specifically, SSD4Rec marks the variable- and long-length item sequences with sequence registers and processes the item representations with bidirectional Structured State Space Duality (SSD) blocks. This not only allows for hardware-aware matrix multiplication but also empowers outstanding capabilities in variable-length and long-range sequence modeling. Extensive evaluations on four benchmark datasets demonstrate that the proposed model achieves state-of-the-art performance while maintaining near-linear scalability with user sequence length. Our code is publicly available at https://github.com/ZhangYifeng1995/SSD4Rec.

Unsupervised Density Neural Representation for CT Metal Artifact Reduction

Emerging unsupervised reconstruction techniques based on implicit neural representation (INR), such as NeRP, CoIL, and SCOPE, have shown unique capabilities in CT linear inverse imaging. In this work, we propose a novel unsupervised density neural representation (Diner) to tackle the challenging problem of CT metal artifacts when scanned objects contain metals. The drastic variation of linear attenuation coefficients (LACs) of metals over X-ray spectra leads to a nonlinear beam hardening effect (BHE) in CT measurements. Recovering CT images from metal-affected measurements therefore poses a complicated nonlinear inverse problem. Existing metal artifact reduction (MAR) techniques mostly formulate the MAR as an image inpainting task, which ignores the energy-induced BHE and produces suboptimal performance. Instead, our Diner introduces an energy-dependent polychromatic CT forward model to the INR framework, addressing the nonlinear nature of the MAR problem. Specifically, we decompose the energy-dependent LACs into energy-independent densities and energy-dependent mass attenuation coefficients (MACs) by fully considering the physical model of X-ray absorption. Using the densities as pivot variables and the MACs as known prior knowledge, the LACs can be accurately reconstructed from the raw measurements. Technically, we represent the unknown density map as an implicit function of coordinates. Combined with a novel differentiable forward model simulating the physical acquisition from the densities to the measurements, our Diner optimizes a multi-layer perception network to approximate the implicit function by minimizing predicted errors between the estimated and real measurements. Experimental results on simulated and real datasets confirm the superiority of our unsupervised Diner against popular supervised techniques in MAR performance and robustness.

Two-stream Convolutional Networks for Multi-frame Face Anti-spoofing

Face anti-spoofing is an important task to protect the security of face recognition. Most of previous work either struggle to capture discriminative and generalizable feature or rely on auxiliary information which is unavailable for most of industrial product. Inspired by the video classification work, we propose an efficient two-stream model to capture the key differences between live and spoof faces, which takes multi-frames and RGB difference as input respectively. Feature pyramid modules with two opposite fusion directions and pyramid pooling modules are applied to enhance feature representation. We evaluate the proposed method on the datasets of Siw, Oulu-NPU, CASIA-MFSD and Replay-Attack. The results show that our model achieves the state-of-the-art results on most of datasets' protocol with much less parameter size.