Ask or search anything...

Retrieval-augmented generation (RAG) incorporates external knowledge into large language models (LLMs), improving their adaptability to downstream tasks and enabling information updates. Surprisingly, recent empirical evidence demonstrates that injecting noise into retrieved relevant documents paradoxically facilitates exploitation of external knowledge and improves generation quality. Although counterintuitive and challenging to apply in practice, this phenomenon enables granular control and rigorous analysis of how LLMs integrate external knowledge. Therefore, in this paper, we intervene on noise injection and establish a layer-specific functional demarcation within the LLM: shallow layers specialize in local context modeling, intermediate layers focus on integrating long-range external factual knowledge, and deeper layers primarily rely on parametric internal knowledge. Building on this insight, we propose Layer Fused Decoding (LFD), a simple decoding strategy that directly combines representations from an intermediate layer with final-layer decoding outputs to fully exploit the external factual knowledge. To identify the optimal intermediate layer, we introduce an internal knowledge score (IKS) criterion that selects the layer with the lowest IKS value in the latter half of layers. Experimental results across multiple benchmarks demonstrate that LFD helps RAG systems more effectively surface retrieved context knowledge with minimal cost.

Video Anomaly Detection (VAD) is an important topic in computer vision. Motivated by the recent advances in self-supervised learning, this paper addresses VAD by solving an intuitive yet challenging pretext task, i.e., spatio-temporal jigsaw puzzles, which is cast as a multi-label fine-grained classification problem. Our method exhibits several advantages over existing works: 1) the spatio-temporal jigsaw puzzles are decoupled in terms of spatial and temporal dimensions, responsible for capturing highly discriminative appearance and motion features, respectively; 2) full permutations are used to provide abundant jigsaw puzzles covering various difficulty levels, allowing the network to distinguish subtle spatio-temporal differences between normal and abnormal events; and 3) the pretext task is tackled in an end-to-end manner without relying on any pre-trained models. Our method outperforms state-of-the-art counterparts on three public benchmarks. Especially on ShanghaiTech Campus, the result is superior to reconstruction and prediction-based methods by a large margin.

The rapid development of online media has significantly facilitated the public's information consumption, knowledge acquisition, and opinion exchange. However, it has also led to more violent conflicts in online discussions. Therefore, controversy detection becomes important for computational and social sciences. Previous research on detection methods has primarily focused on larger datasets and more complex computational models but has rarely examined the underlying mechanisms of conflict, particularly the psychological motivations behind them. In this paper, we present evidence that conflicting posts tend to have a high proportion of "ascending gradient of likes", i.e., replies get more likes than comments. Additionally, there is a gradient in the number of replies between the neighboring tiers as well. We develop two new gradient features and demonstrate the common enhancement effect of our features in terms of controversy detection models. Further, multiple evaluation algorithms are used to compare structural, interactive, and textual features with the new features across multiple Chinese and English media. The results show that it is a general case that gradient features are significantly different in terms of controversy and are more important than other features. More thoroughly, we discuss the mechanism by which the ascending gradient emerges, suggesting that the case is related to the "backfire effect" in ideological conflicts that have received recent attention. The features formed by the psychological mechanism also show excellent detection performance in application scenarios where only a few hot information or early information are considered. Our findings can provide a new perspective for online conflict behavior analysis and early detection.