A Survey on Large Language Model Acceleration based on KV Cache Management

Large Language Models (LLMs) have revolutionized a wide range of domains such as natural language processing, computer vision, and multi-modal tasks due to their ability to comprehend context and perform logical reasoning. However, the computational and memory demands of LLMs, particularly during inference, pose significant challenges when scaling them to real-world, long-context, and real-time applications. Key-Value (KV) cache management has emerged as a critical optimization technique for accelerating LLM inference by reducing redundant computations and improving memory utilization. This survey provides a comprehensive overview of KV cache management strategies for LLM acceleration, categorizing them into token-level, model-level, and system-level optimizations. Token-level strategies include KV cache selection, budget allocation, merging, quantization, and low-rank decomposition, while model-level optimizations focus on architectural innovations and attention mechanisms to enhance KV reuse. System-level approaches address memory management, scheduling, and hardware-aware designs to improve efficiency across diverse computing environments. Additionally, the survey provides an overview of both text and multimodal datasets and benchmarks used to evaluate these strategies. By presenting detailed taxonomies and comparative analyses, this work aims to offer useful insights for researchers and practitioners to support the development of efficient and scalable KV cache management techniques, contributing to the practical deployment of LLMs in real-world applications. The curated paper list for KV cache management is in: \href{https://github.com/TreeAI-Lab/Awesome-KV-Cache-Management}{https://github.com/TreeAI-Lab/Awesome-KV-Cache-Management}.

Cross-Attention Graph Neural Networks for Inferring Gene Regulatory Networks with Skewed Degree Distribution

Inferencing Gene Regulatory Networks (GRNs) from gene expression data is a pivotal challenge in systems biology, and several innovative computational methods have been introduced. However, most of these studies have not considered the skewed degree distribution of genes. Specifically, some genes may regulate multiple target genes while some genes may be regulated by multiple regulator genes. Such a skewed degree distribution issue significantly complicates the application of directed graph embedding methods. To tackle this issue, we propose the Cross-Attention Complex Dual Graph Embedding Model (XATGRN). Our XATGRN employs a cross-attention mechanism to effectively capture intricate gene interactions from gene expression profiles. Additionally, it uses a Dual Complex Graph Embedding approach to manage the skewed degree distribution, thereby ensuring precise prediction of regulatory relationships and their directionality. Our model consistently outperforms existing state-of-the-art methods across various datasets, underscoring its efficacy in elucidating complex gene regulatory mechanisms. Our codes used in this paper are publicly available at: https://github.com/kikixiong/XATGRN.

An Investigation on the Potential of KAN in Speech Enhancement

High-fidelity speech enhancement often requires sophisticated modeling to capture intricate, multiscale patterns. Standard activation functions, while introducing nonlinearity, lack the flexibility to fully address this complexity. Kolmogorov-Arnold Networks (KAN), an emerging methodology that employs learnable activation functions on graph edges, present a promising alternative. This work investigates two novel KAN variants based on rational and radial basis functions for speech enhancement. We integrate the rational variant into the 1D CNN blocks of Demucs and the GRU-Transformer blocks of MP-SENet, while the radial variant is adapted to the 2D CNN-based decoders of MP-SENet. Experiments on the VoiceBank-DEMAND dataset show that replacing standard activations with KAN-based activations improves speech quality across both the time-domain and time-frequency domain methods with minimal impact on model size and FLOP, underscoring KAN's potential to improve speech enhancement models.

Demystifying Workload Imbalances in Large Transformer Model Training over Variable-length Sequences

To optimize large Transformer model training, efficient parallel computing and advanced data management are essential. However, current methods often assume a stable and uniform training workload, neglecting imbalances in data sampling and packing that can impede performance. Specifically, data sampling imbalance arises from uneven sequence length distribution of the training data, while data packing imbalance stems from the discrepancy between the linear memory complexity and quadratic time complexity of the attention mechanism. To address these imbalance issues, we develop Hydraulis, which jointly optimizes the parallel strategies and data assignment. For one thing, we introduce large model training with dynamic heterogeneous parallel strategies in response to the sequence length variations within and across training iterations. For another, we devise a two-stage data assignment approach, which strikes a good balance in terms of the training workloads both within and across model replicas. Empirical results demonstrate that Hydraulis outperforms existing systems by 1.32-2.66 times.

Reconstruction of boosted and resolved multi-Higgs-boson events with symmetry-preserving attention networks

The production of multiple Higgs bosons at the CERN LHC provides a direct way to measure the trilinear and quartic Higgs self-interaction strengths as well as potential access to beyond the standard model effects that can enhance production at large transverse momentum $p_{\mathrm{T}}$. The largest event fraction arises from the fully hadronic final state in which every Higgs boson decays to a bottom quark-antiquark pair ($b\bar{b}$). This introduces a combinatorial challenge known as the \emph{jet assignment problem}: assigning jets to sets representing Higgs boson candidates. Symmetry-preserving attention networks (SPA-Nets) have been been developed to address this challenge. However, the complexity of jet assignment increases when simultaneously considering both $H\rightarrow b\bar{b}$ reconstruction possibilities, i.e., two "resolved" small-radius jets each containing a shower initiated by a $b$-quark or one "boosted" large-radius jet containing a merged shower initiated by a $b\bar{b}$ pair. The latter improves the reconstruction efficiency at high $p_{\mathrm{T}}$. In this work, we introduce a generalization to the SPA-Net approach to simultaneously consider both boosted and resolved reconstruction possibilities and unambiguously interpret an event as "fully resolved'', "fully boosted", or in between. We report the performance of baseline methods, the original SPA-Net approach, and our generalized version on nonresonant $HH$ and $HHH$ production at the LHC. Considering both boosted and resolved topologies, our SPA-Net approach increases the Higgs boson reconstruction purity by 57--62\% and the efficiency by 23--38\% compared to the baseline method depending on the final state.

Learning Symmetry-Independent Jet Representations via Jet-Based Joint Embedding Predictive Architecture

In high energy physics, self-supervised learning (SSL) methods have the potential to aid in the creation of machine learning models without the need for labeled datasets for a variety of tasks, including those related to jets -- narrow sprays of particles produced by quarks and gluons in high energy particle collisions. This study introduces an approach to learning jet representations without hand-crafted augmentations using a jet-based joint embedding predictive architecture (J-JEPA), which aims to predict various physical targets from an informative context. As our method does not require hand-crafted augmentation like other common SSL techniques, J-JEPA avoids introducing biases that could harm downstream tasks. Since different tasks generally require invariance under different augmentations, this training without hand-crafted augmentation enables versatile applications, offering a pathway toward a cross-task foundation model. We finetune the representations learned by J-JEPA for jet tagging and benchmark them against task-specific representations.

Noro: A Noise-Robust One-shot Voice Conversion System with Hidden Speaker Representation Capabilities

One-shot voice conversion (VC) aims to alter the timbre of speech from a source speaker to match that of a target speaker using just a single reference speech from the target, while preserving the semantic content of the original source speech. Despite advancements in one-shot VC, its effectiveness decreases in real-world scenarios where reference speeches, often sourced from the internet, contain various disturbances like background noise. To address this issue, we introduce Noro, a Noise Robust One-shot VC system. Noro features innovative components tailored for VC using noisy reference speeches, including a dual-branch reference encoding module and a noise-agnostic contrastive speaker loss. Experimental results demonstrate that Noro outperforms our baseline system in both clean and noisy scenarios, highlighting its efficacy for real-world applications. Additionally, we investigate the hidden speaker representation capabilities of our baseline system by repurposing its reference encoder as a speaker encoder. The results shows that it is competitive with several advanced self-supervised learning models for speaker representation under the SUPERB settings, highlighting the potential for advancing speaker representation learning through one-shot VC task.

Subgraph Aggregation for Out-of-Distribution Generalization on Graphs

Out-of-distribution (OOD) generalization in Graph Neural Networks (GNNs) has gained significant attention due to its critical importance in graph-based predictions in real-world scenarios. Existing methods primarily focus on extracting a single causal subgraph from the input graph to achieve generalizable predictions. However, relying on a single subgraph can lead to susceptibility to spurious correlations and is insufficient for learning invariant patterns behind graph data. Moreover, in many real-world applications, such as molecular property prediction, multiple critical subgraphs may influence the target label property. To address these challenges, we propose a novel framework, SubGraph Aggregation (SuGAr), designed to learn a diverse set of subgraphs that are crucial for OOD generalization on graphs. Specifically, SuGAr employs a tailored subgraph sampler and diversity regularizer to extract a diverse set of invariant subgraphs. These invariant subgraphs are then aggregated by averaging their representations, which enriches the subgraph signals and enhances coverage of the underlying causal structures, thereby improving OOD generalization. Extensive experiments on both synthetic and real-world datasets demonstrate that \ours outperforms state-of-the-art methods, achieving up to a 24% improvement in OOD generalization on graphs. To the best of our knowledge, this is the first work to study graph OOD generalization by learning multiple invariant subgraphs.

Decentralized Hybrid Precoding for Massive MU-MIMO ISAC

Integrated sensing and communication (ISAC) is a very promising technology designed to provide both high rate communication capabilities and sensing capabilities. However, in Massive Multi User Multiple-Input Multiple-Output (Massive MU MIMO-ISAC) systems, the dense user access creates a serious multi-user interference (MUI) problem, leading to degradation of communication performance. To alleviate this problem, we propose a decentralized baseband processing (DBP) precoding method. We first model the MUI of dense user scenarios with minimizing Cramer-Rao bound (CRB) as an objective function.Hybrid precoding is an attractive ISAC technique, and hybrid precoding using Partially Connected Structures (PCS) can effectively reduce hardware cost and power consumption. We mitigate the MUI between dense users based on ThomlinsonHarashima Precoding (THP). We demonstrate the effectiveness of the proposed method through simulation experiments. Compared with the existing methods, it can effectively improve the communication data rates and energy efficiency in dense user access scenario, and reduce the hardware complexity of Massive MU MIMO-ISAC systems. The experimental results demonstrate the usefulness of our method for improving the MUI problem in ISAC systems for dense user access scenarios.

MixPolyp: Integrating Mask, Box and Scribble Supervision for Enhanced Polyp Segmentation

Limited by the expensive labeling, polyp segmentation models are plagued by data shortages. To tackle this, we propose the mixed supervised polyp segmentation paradigm (MixPolyp). Unlike traditional models relying on a single type of annotation, MixPolyp combines diverse annotation types (mask, box, and scribble) within a single model, thereby expanding the range of available data and reducing labeling costs. To achieve this, MixPolyp introduces three novel supervision losses to handle various annotations: Subspace Projection loss (L_SP), Binary Minimum Entropy loss (L_BME), and Linear Regularization loss (L_LR). For box annotations, L_SP eliminates shape inconsistencies between the prediction and the supervision. For scribble annotations, L_BME provides supervision for unlabeled pixels through minimum entropy constraint, thereby alleviating supervision sparsity. Furthermore, L_LR provides dense supervision by enforcing consistency among the predictions, thus reducing the non-uniqueness. These losses are independent of the model structure, making them generally applicable. They are used only during training, adding no computational cost during inference. Extensive experiments on five datasets demonstrate MixPolyp's effectiveness.