Retrieval Augmented Generation (RAG) has gained widespread adoption owing to its capacity to empower large language models (LLMs) to integrate external knowledge. However, existing RAG frameworks are primarily designed for text-based LLMs and rely on Automatic Speech Recognition to process speech input, which discards crucial audio information, risks transcription errors, and increases computational overhead. Therefore, we introduce WavRAG, the first retrieval augmented generation framework with native, end-to-end audio support. WavRAG offers two key features: 1) Bypassing ASR, WavRAG directly processes raw audio for both embedding and retrieval. 2) WavRAG integrates audio and text into a unified knowledge representation. Specifically, we propose the WavRetriever to facilitate the retrieval from a text-audio hybrid knowledge base, and further enhance the in-context capabilities of spoken dialogue models through the integration of chain-of-thought reasoning. In comparison to state-of-the-art ASR-Text RAG pipelines, WavRAG achieves comparable retrieval performance while delivering a 10x acceleration. Furthermore, WavRAG's unique text-audio hybrid retrieval capability extends the boundaries of RAG to the audio modality.

The expanding context windows in large language models (LLMs) have greatly enhanced their capabilities in various applications, but they also introduce significant challenges in maintaining low latency, particularly in Time to First Token (TTFT). This paper identifies that the sharp rise in TTFT as context length increases is predominantly driven by queuing delays, which are caused by the growing demands for GPU Key-Value (KV) cache allocation clashing with the limited availability of KV cache blocks. To address this issue, we propose LayerKV, a simple yet effective plug-in method that effectively reduces TTFT without requiring additional hardware or compromising output performance, while seamlessly integrating with existing parallelism strategies and scheduling techniques. Specifically, LayerKV introduces layer-wise KV block allocation, management, and offloading for fine-grained control over system memory, coupled with an SLO-aware scheduler to optimize overall Service Level Objectives (SLOs). Comprehensive evaluations on representative models, ranging from 7B to 70B parameters, across various GPU configurations, demonstrate that LayerKV improves TTFT latency up to 69x and reduces SLO violation rates by 28.7%, significantly enhancing the user experience.

Sequential modeling has demonstrated remarkable capabilities in offline reinforcement learning (RL), with Decision Transformer (DT) being one of the most notable representatives, achieving significant success. However, RL trajectories possess unique properties to be distinguished from the conventional sequence (e.g., text or audio): (1) local correlation, where the next states in RL are theoretically determined solely by current states and actions based on the Markov Decision Process (MDP), and (2) global correlation, where each step's features are related to long-term historical information due to the time-continuous nature of trajectories. In this paper, we propose a novel action sequence predictor, named Mamba Decision Maker (MambaDM), where Mamba is expected to be a promising alternative for sequence modeling paradigms, owing to its efficient modeling of multi-scale dependencies. In particular, we introduce a novel mixer module that proficiently extracts and integrates both global and local features of the input sequence, effectively capturing interrelationships in RL datasets. Extensive experiments demonstrate that MambaDM achieves state-of-the-art performance in Atari and OpenAI Gym datasets. Furthermore, we empirically investigate the scaling laws of MambaDM, finding that increasing model size does not bring performance improvement, but scaling the dataset amount by 2x for MambaDM can obtain up to 33.7% score improvement on Atari dataset. This paper delves into the sequence modeling capabilities of MambaDM in the RL domain, paving the way for future advancements in robust and efficient decision-making systems. Our code will be available at https://github.com/AndyCao1125/MambaDM.

Spiking Transformers have gained considerable attention because they achieve both the energy efficiency of Spiking Neural Networks (SNNs) and the high capacity of Transformers. However, the existing Spiking Transformer architectures, derived from Artificial Neural Networks (ANNs), exhibit a notable architectural gap, resulting in suboptimal performance compared to their ANN counterparts. Manually discovering optimal architectures is time-consuming. To address these limitations, we introduce AutoST, a training-free NAS method for Spiking Transformers, to rapidly identify high-performance Spiking Transformer architectures. Unlike existing training-free NAS methods, which struggle with the non-differentiability and high sparsity inherent in SNNs, we propose to utilize Floating-Point Operations (FLOPs) as a performance metric, which is independent of model computations and training dynamics, leading to a stronger correlation with performance. Our extensive experiments show that AutoST models outperform state-of-the-art manually or automatically designed SNN architectures on static and neuromorphic datasets. Full code, model, and data are released for reproduction.

DBSCAN has been widely used in density-based clustering algorithms. However, with the increasing demand for Multi-density clustering, previous traditional DSBCAN can not have good clustering results on Multi-density datasets. In order to address this problem, an adaptive Multi-density DBSCAN algorithm (AMD-DBSCAN) is proposed in this paper. An improved parameter adaptation method is proposed in AMD-DBSCAN to search for multiple parameter pairs (i.e., Eps and MinPts), which are the key parameters to determine the clustering results and performance, therefore allowing the model to be applied to Multi-density datasets. Moreover, only one hyperparameter is required for AMD-DBSCAN to avoid the complicated repetitive initialization operations. Furthermore, the variance of the number of neighbors (VNN) is proposed to measure the difference in density between each cluster. The experimental results show that our AMD-DBSCAN reduces execution time by an average of 75% due to lower algorithm complexity compared with the traditional adaptive algorithm. In addition, AMD-DBSCAN improves accuracy by 24.7% on average over the state-of-the-art design on Multi-density datasets of extremely variable density, while having no performance loss in Single-density scenarios. Our code and datasets are available at https://github.com/AlexandreWANG915/AMD-DBSCAN.

Inpatient falls are a serious safety issue in hospitals and healthcare facilities. Recent advances in video analytics for patient monitoring provide a non-intrusive avenue to reduce this risk through continuous activity monitoring. However, in-bed fall risk assessment systems have received less attention in the literature. The majority of prior studies have focused on fall event detection, and do not consider the circumstances that may indicate an imminent inpatient fall. Here, we propose a video-based system that can monitor the risk of a patient falling, and alert staff of unsafe behaviour to help prevent falls before they occur. We propose an approach that leverages recent advances in human localisation and skeleton pose estimation to extract spatial features from video frames recorded in a simulated environment. We demonstrate that body positions can be effectively recognised and provide useful evidence for fall risk assessment. This work highlights the benefits of video-based models for analysing behaviours of interest, and demonstrates how such a system could enable sufficient lead time for healthcare professionals to respond and address patient needs, which is necessary for the development of fall intervention programs.