AI conference papers

AI conference paper library

Search papers from 30+ top conferences, including NeurIPS, ICLR, ICML, CVPR, ACL and AAAI. Filter by keyword, then open any paper for its abstract, authors and links.

arXiv

Error Certificates for KV-Cache Eviction via Randomized Design

Peng Xie, Peng Xie Amr Alanwar

Deterministic KV-cache eviction keeps the top-$k$ tokens under an importance score and deletes the rest, and after the deletion the serving system cannot know what the eviction cost it on the current query. We replace the deterministic tail with Poisson sampling at known inclusion probabilities, which makes the eviction error identifiable and turns a survey-sampling variance estimator over the retained set into a per-step error certificate at one extra scalar per retained token. On a thirty-turn assistant compressed to a 10\% cache budget, the certificate-gated system answers 0.97 of recall questions against 0.09 for top-$k$, and for facts stated 26 to 30 turns earlier it recalls 97\% against 2\%. We prove that no estimator computable from the information a deterministic scheme retains is consistent for its own eviction error: evicted values can be altered so that everything retained is unchanged while the true attention-output error grows without bound. Under the Poisson design the certificate covers the realized attention error in 96.9--97.7\% of 12{,}096 replay cells and in 98.1--99.7\% on twelve further architectures. Randomization buys attribution, not prediction: a pre-registered study on LongBench at 6k and 16k tokens (about 74{,}000 generations) finds question-aware eviction at 25--50\% budgets nearly free and output log-probability the better failure predictor, while the certificate answers the question confidence cannot, separating eviction-induced from inherent failures at AUC 0.65--0.75 against 0.47--0.54, and schedules recomputation at 1.7--1.8 times the gain of random gating. On real long-term conversations the gated system returns the full-cache score inside the heavy-damage regime, and the rule that triggers it is the same across five model families.

arXiv

Capital Markets LLM Reliability Score (CM-LRS): From Plausible to Bankable

Prerit Ahuja

In capital-markets workflows the question is rarely whether a large language model can produce a fluent draft, but whether the draft is bankable: defensible in front of a counter-party or a regulator, with the documents in hand. Existing methods address parts of that gap: open-domain QA benchmarks reward surface accuracy, and finance benchmarks (FinanceBench, FinQA, ConvFinQA) advance document-grounded and numerical QA but evaluate at the question-answer layer rather than the workflow outputs practitioners defend. We introduce CM-LRS, a Capital Markets LLM Reliability Score, evaluating outputs at the workflow-output layer across seven dimensions: factual accuracy, evidence traceability, numerical consistency, workflow completeness, source discipline, decision usefulness, and reviewability/auditability. Each is scored 0-5 against a rubric anchored on signals reviewers in regulated settings use; the aggregate is tunable to the workflow. We demonstrate CM-LRS on five workflows (DCM transaction-terms extraction, precedent retrieval, issuer profile synthesis, M&A transaction-comparable reasoning, ECM transaction-terms extraction) over public SEC EDGAR filings, a public UK takeover release, and fictional synthetic supplements, scoring four models against four independent LLM judges spanning three model families. Three findings. First, the frontier closed-source models cluster within 0.22 points on four-judge averaged CM-LRS (Sonnet 4.6 = 4.31, Opus 4.7 = 4.30, GPT-5.5 = 4.09); all four judges place the open-weights baseline (Llama 3.3 70B = 3.15) last. Second, that gap concentrates on retrieval (2.23) and synthesis (2.15), not extraction (0.84). Third, Decision Usefulness shows the widest cross-model dispersion of any dimension (4.0 points on issuer profiling) and top-tier inter-judge agreement (mean r = 0.52). Plausibility is cheap. Bankability is the bar.

arXiv

VibeVoice-ASR-BitNet Technical Report

Songchen Xu, Ting Song, Shaohan Huang, Zhiliang Peng, Yan Xia, Yujie Tu, Xin Huang, Xun Wu, Wenhui Wang, Yaoyao Chang, Jianwei Yu, Li Dong, Furu Wei

We present VibeVoice-ASR-BitNet, a compressed variant of VibeVoice-ASR optimized for real-time inference on edge CPUs. We apply heterogeneous quantization tailored to the computational characteristics of each stage: the VAE acoustic tokenizer uses full-pipeline INT8 quantization (I8_S) with kernel fusion and SIMD optimization, while the autoregressive language model adopts BitNet-style ternary weights (I2_S). To preserve accuracy under aggressive compression, we employ a progressive quantization-aware training strategy. For inference, we implement custom SIMD kernels and fused operators within the ggml framework targeting both ARM and x86 platforms, achieving real-time recognition (RTF < 1) on low-thread-count CPUs. VibeVoice-ASR-BitNet is 1.6--2.3x faster than Whisper.cpp at comparable model sizes (~1.6 GB), with only modest accuracy degradation compared to the FP16 baseline.

arXiv

An MLIR-Based Compilation Method for Large Language Models

Pengchao Hu, Zhibin Xin, Yifan Chen, Yangyang Zhou, Liang Wang, Xin Zhang

Large Language Models (LLMs) have become the dominant workload on modern AI accelerators, yet deploying them on specialized hardware still faces two core challenges: how to import a trained model into a compiler-friendly intermediate representation, and how to efficiently schedule the autoregressive inference loop under limited on-chip memory. This paper presents an MLIR (Multi-Level Intermediate Representation) based compilation method for large language models, illustrated using two dialects of operators, TopOp and TpuOp. TopOp serves as a high-level graph dialect that is independent of both the source framework and the target chip, and is responsible for expressing model semantics; TpuOp serves as the target hardware dialect, carrying chip-related decisions such as quantization, layer groups, and memory layout. A model is first represented as TopOp, then lowered layer by layer to TpuOp, and finally a deployable binary is generated. In addition, each Transformer layer is split into three stages for static compilation: prefill, prefill_kv (prefill with historical key-value cache), and decode, so as to accommodate the different computational characteristics of prompt-parallel processing and per-token generation. The method has been implemented in the TPU-MLIR compiler {https://github.com/sophgo/tpu-mlir} and the LLM-TPU deployment project {https://github.com/sophgo/LLM-TPU}, supporting a variety of generative models including the Qwen, Llama, InternVL, and MiniCPM-V series, as well as multiple quantization and deployment forms such as GPTQ, AWQ, and AutoRound.

arXiv

The One-Word Census: Answer-Choice Conformity Across 44 Language Models

Tapan Parikh

When a language model must pick one answer from a large space of equally valid options, which does it pick -- and how often is it the same answer every other model picks? Asked to "pick a word -- any word," 44 models chose "serendipity" 41% of the time. We characterize this convergence with a deliberately minimal instrument: 31 single-turn prompts, each naming a category with many valid one-word answers ("Name a tree."), asked four times per model with no system prompt. Analysis is exact-match on normalized tokens -- no embeddings, no judge -- at about a dollar per model. That models converge is well documented; our contribution is the instrument itself -- the One-Word Census -- and what it reveals about the structure of the convergence. We score each model by answer-choice surprisal: the average $-\log2$ probability of its answers under the pooled answers of all other models, leave-one-out. Convergence is extreme -- in 7 of 31 categories one answer takes over 80% of all answers -- yet conformity varies more than fourfold across models, and the variation is structured. Persona- and community-tuned models are the most divergent; the newest mainline flagships are the most conformist, producing almost no answer no other model gave. Within four lineages (Claude, GPT, Qwen, Grok) conformity rises with each generation -- but reverses for the latest flagship Claude and GPT models, a possible early signal of repositioning at the top tier. Rankings are robust to roster composition (leave-one-family-out rho = 0.985). Against human category-production norms, the field is more concentrated than people in 18 of 20 shared categories. All prompts, transcripts, and code are public.

arXiv

Neural Machine Translation for Low-Resource Tangkhul--English

Chormi Zimik Vashai, Agniva Maiti

We present a study on low-resource machine translation for the Tangkhul-English (nmf-en) language pair. Tangkhul is a severely under-resourced Tibeto-Burman language spoken primarily in Manipur, India, with virtually no prior natural language processing infrastructure. We describe two systems: (1) a primary system based on ByT5-large fine-tuned on 38,336 Tangkhul-English parallel sentence pairs, and (2) a contrastive system based on mT5-small fine-tuned on the same corpus. Our primary ByT5-large system achieves a corpus BLEU score of 39.97, chrF++ of 58.07, BERTScore F1 of 0.8104, and COMET (wmt22-comet-da) of 0.7302 on a held-out test set of 3,856 sentences. We further discuss the orthographic challenges specific to Tangkhul's Latin-script diacritics, the domain bias of our training corpus (which comprises biblical text, stories, and conversational data), and avenues for future improvement through data diversification and domain adaptation.

arXiv

daVinci-kernel: Co-Evolving Skill Selection, Summarization, and Utilization via RL for GPU Kernel Optimization

Dayuan Fu, Mohan Jiang, Tongyu Wang, Dian Yang, Jiarui Hu, Liming Liu, Jinlong Hou, Pengfei Liu

GPU kernel optimization represents a paradigm where functional correctness is assumed and execution efficiency is the objective. We present daVinci-kernel, a reinforcement learning framework that couples skill discovery with skill exploitation through a dynamically evolving skill library. daVinci-kernel jointly trains three agents sharing one LLM backbone: a Skill Selection Agent that retrieves relevant techniques via BM25 and LLM reranking, a Policy Agent that generates multi-turn CUDA/Triton kernels conditioned on selected skills, and a Skill Summary Agent that distills successful rollouts into reusable skills. Candidate skills are added only after execution-based verification confirms reproducible speedups. All three agents share a single LLM backbone, are initialized via a structured SFT cold start on diversity-filtered data, and are then jointly optimized end-to-end with multi-turn REINFORCE and per-agent advantage estimation. On KernelBench, daVinci-kernel-14B achieves 37.2%, 70.6%, and 32.2% on Level 1, Level 2, and Level 3 under the Fast$_1$ threshold, outperforming the strongest prior RL-trained model, Dr\. Kernel-14B.

arXiv

Algorithmic Blindness in Large Language Models: A Calibration Study of Performance Prediction

Sohan Venkatesh, Ashish Mahendran Kurapath, Tejas Melkote

Large language models (LLMs) demonstrate remarkable breadth of knowledge, yet their ability to reason about computational processes remains poorly understood. Closing this gap matters for practitioners who rely on LLMs to guide algorithm selection and deployment. We address this limitation using causal discovery as a testbed and evaluate eight frontier LLMs against ground truth derived from algorithm executions. We find systematic, near-total failure across models. The predicted ranges are far wider than true confidence intervals yet still fail to contain the true algorithmic mean in most cases. Most models perform worse than random guessing. The best model's marginal improvement points to benchmark memorization rather than principled reasoning. We term this failure algorithmic blindness and argue it reflects a fundamental gap between declarative knowledge about algorithms and calibrated procedural prediction.

arXiv

Does Faithfulness-Guided Alignment Hurt Accuracy? Unlocking Accurate and Faithful Post-Retrieval Reasoning

Yu Liu, Wenxiao Zhang, Diandian Guo, Cong Cao, Fangfang Yuan, Qiang Sun, Yanbing Liu, Jin B. Hong, Zhiyuan Ma

Retrieval-augmented generation (RAG) can achieve strong answer accuracy on multi-hop questions, but outcome-level rewards often leave reasoning traces weakly grounded and difficult to audit. Under noisy retrieval, models may exhibit right-answer-wrong-reason failures, where the final answer is correct but the supporting rationale exploits shortcuts or unsupported evidence. We therefore ask whether faithfulness-guided alignment hurts answer accuracy in post-retrieval reasoning. To study this question, we propose CRAFT (Calibrated Reasoning with Answer-Faithful Traces), a reinforcement learning framework for the response-generation stage of retrieval-augmented multi-hop question answering. CRAFT trains models to produce structured reasoning traces with configurable auditability, while combining deterministic rewards for format compliance, answer correctness, and citation validity with a judge-based reward for semantic faithfulness. Experiments across model scales and benchmarks show that CRAFT unlocks task-specific reasoning capacity from 1.5B upward, improving both answer accuracy and Faithfulness; at 0.5B, performance remains sharply template-dependent. At 7B, CRAFT improves Faithfulness over the Base model in all evaluated settings and remains competitive with strong closed-source models. Code is available at https://github.com/Ameame1/CRAFT.

arXiv

SERM: Self-Evolving Relevance Model with Agent-Driven Learning from Massive Query Streams

Chenglong Wang, Canjia Li, Xingzhao Zhu, Yifu Huo, Huiyu Wang, Weixiong Lin, Yun Yang, Qiaozhi He, Tianhua Zhou, Xiaojia Chang, Jingbo Zhu, Tong Xiao

Due to the dynamically evolving nature of real-world query streams, relevance models struggle to generalize to practical search scenarios. A sophisticated solution is self-evolution techniques. However, in large-scale industrial settings with massive query streams, this technique faces two challenges: (1) informative samples are often sparse and difficult to identify, and (2) pseudo-labels generated by the current model could be unreliable. To address these challenges, in this work, we propose a Self-Evolving Relevance Model approach (SERM), which comprises two complementary multi-agent modules: a multi-agent sample miner, designed to detect distributional shifts and identify informative training samples, and a multi-agent relevance annotator, which provides reliable labels through a two-level agreement framework. We evaluate SERM in a large-scale industrial setting, which serves billions of user requests daily. Experimental results demonstrate that SERM can achieve significant performance gains through iterative self-evolution, as validated by extensive offline multilingual evaluations and online testing.

arXiv

LEDOM: Reverse Language Model

Xunjian Yin, Sitao Cheng, Yuxi Xie, Xinyu Hu, Li Lin, Xinyi Wang, Liangming Pan, William Yang Wang, Xiaojun Wan

Autoregressive language models are trained exclusively left-to-right. We explore the complementary factorization, training right-to-left at scale, and ask what reasoning patterns emerge when a model conditions on future context to predict the past. We train LEDOM, an open-source purely reverse autoregressive language model (2B/7B parameters, 435B tokens), and find it develops capabilities distinct from forward models, including abductive inference, question synthesis, and natural resolution of the reversal curse. We then explore one application of the reverse model: combining forward likelihood $P(y \mid x)$ with reverse posterior $P(x \mid y)$ through noisy channel duality. We propose Reverse Reward, which reranks forward outputs using reverse posterior estimates, and prove that bidirectional scoring penalizes hallucinated reasoning chains whose backward reconstruction degrades. Reverse Reward yields gains of up to 6.6\% on AIME 2024 and 15\% on AMC 2023 across multiple strong baselines. We release all models, code, and data here: https://github.com/Arvid-pku/LEDOM.

arXiv

Flick: Few Labels Text Classification using K-Aware Intermediate Learning in Multi-Task Low-Resource Languages

Ali Almutairi, Abdullah Alsuhaibani, Shoaib Jameel, Aditya Joshi, Gelareh Mohammadi, Imran Razzak

Training deep learning networks with minimal supervision has gained significant research attention due to its potential to reduce reliance on extensive labelled data. While self-training methods have proven effective in semi-supervised learning, they remain vulnerable to errors from noisy pseudo labels. Moreover, most recent approaches to the few-label classification problem are either designed for resource-rich languages such as English or involve complex cascading models that are prone to overfitting. To address the persistent challenge of few-label text classification in truly low-resource linguistic contexts, where existing methods often struggle with noisy pseudo-labels and domain adaptation, we propose Flick. Unlike prior methods that rely on generic multi-cluster pseudo-labelling or complex cascading architectures, Flick leverages the fundamental insight that distilling high-confidence pseudo-labels from a broader set of initial clusters can dramatically improve pseudo-label quality, particularly for linguistically diverse, low-resource settings. Flick introduces a novel pseudo-label refinement component, a departure from traditional pseudo-labelling strategies by identifying and leveraging top-performing pseudo-label clusters. This component specifically learns to distil highly reliable pseudo-labels from an initial broad set by focusing on single-cluster cohesion and leveraging an adaptive top-k selection mechanism. This targeted refinement process is crucial for mitigating the propagation of errors inherent in low-resource data, allowing for robust fine-tuning of pre-trained language models with only a handful of true labels. We demonstrate Flick's efficacy across 14 diverse datasets, encompassing challenging low-resource languages such as Arabic, Urdu, and Setswana, alongside English, showcasing its superior performance and adaptability.

arXiv

SAFE-MEME: Structured Reasoning Framework for Robust Hate Speech Detection in Memes

Palash Nandi, Shivam Sharma, Tanmoy Chakraborty

Memes act as cryptic tools for sharing sensitive ideas, often requiring contextual knowledge to interpret them correctly. It makes multimodal meme moderation difficult, as existing work either lacks high-quality datasets for nuanced hate categories or relies on low-quality social media visuals. Here, we curate two novel multimodal hate speech datasets comprising English memes - MHS and MHS-Con, which capture fine-grained hateful abstractions in regular and confounding scenarios, respectively. We benchmark these datasets against several competing baselines. Furthermore, we introduce SAFE-MEME (Structured reAsoning FramEwork) with its two variants: a novel multimodal Chain-of-Thought based framework employing Q&A-style reasoning (SAFE-MEME-QA) and a hierarchical categorization (SAFE-MEME-H) to enable robust hate speech detection in memes. SAFE-MEME-QA outperforms the strongest open-source baseline model, showing an improvement of 2% on MHS and 4.7% on MHS-Con and closely follows the closed-source models, GPT-4o and Gemini 2.5. In contrast, SAFE-MEME-H surpasses SAFE-MEME-QA, marking a 3% improvement over the best open-source baseline, while achieving performance comparable to GPT-4o only on MHS. We show that fine-tuning a single-layer adapter within SAFE-MEME-H outperforms fully fine-tuned models in regular fine-grained hateful meme detection. However, the fully fine-tuning approach with a Q&A setup is more effective for handling confounding cases. We also systematically examine the error cases, offering valuable insights into the robustness and limitations of the proposed structured reasoning framework for analyzing hateful memes.

arXiv

LIBMoE: A Library for comprehensive benchmarking Mixture of Experts in Large Language Models

Nam V. Nguyen, Thong T. Doan, Luong Tran, Van Nguyen, Quang Pham

Mixture of experts (MoE) architectures have become a cornerstone for scaling up and are a key component in most large language models such as GPT-OSS, DeepSeek-V3, Llama-4, and Gemini-2.5. However, systematic research on MoE remains severely constrained by the prohibitive computational costs of training and evaluation, restricting large-scale studies accessible to most researchers. We introduce LibMoE, a unified framework for reproducible, efficient, and extensible MoE research that supports both pretraining and sparse-upcycling regimes. Beyond unified implementations, the framework provides transparent analytical tools for probing routing and expert dynamics. Leveraging this foundation, we conduct a comprehensive analysis along three dimensions: (i) routing dynamics, covering expert selection patterns, routing stability and optimality, and how routing entropy reveals task specialization and expert diversity; (ii) the effect of lightweight initialization on load balancing, demonstrating how subtle changes in router initialization shape early expert utilization; and (iii) training regime differences, revealing how sparse upcycling and full pretraining exhibit distinct routing patterns and stability profiles. By lowering the barrier to entry and standardizing evaluation, along with our comprehensive analysis, LibMoE broadens access to MoE research and establishes a reliable benchmark to guide future innovations. GitHub: \href{https://github.com/Fsoft-AIC/LibMoE}{https://github.com/Fsoft-AIC/LibMoE}.

arXiv

No Optimal Language Set Exists for Multilingual Instruction Tuning: Insights from a Linguistically-Informed Study

Gürkan Soykan, Gözde Gül Şahin

Multilingual instruction tuning (MIT) is challenged by the curse of multilinguality, data scarcity, and high computational cost. A natural hypothesis is that carefully selecting a linguistically diverse set of languages yields universally better models. We test this systematically by evaluating linguistically-informed selection strategies---based on typological, geographical, semantic, and learned features---against random baselines across three model families (mGPT, mT5-xl, BLOOM) and five multilingual benchmarks. Our key negative finding is that no universal language selection strategy emerges in our fixed-budget setting: performance is strongly task- and model-dependent, and adding more languages beyond a modest threshold triggers degradation consistent with the curse of multilinguality. We discuss implications for MIT data curation and the pitfalls of benchmark-averaged evaluation. All resources are publicly available at https://github.com/GGLAB-KU/ling-informed-mit.

arXiv

Speed Reading Tool Powered by Artificial Intelligence for Students with ADHD, Dyslexia, and Short Attention Span

Megat Irfan Zackry Bin Ismail, Ahmad Nazran bin Yusri, Muhammad Hafizzul Bin Abdul Manap, Muhammad Muizzuddin Bin Kamarozaman

This paper presents an artificial intelligence tool designed to assist students with dyslexia, ADHD, and short attention spans in processing text-based information more efficiently. The proposed solution addresses both cognitive and visual reading barriers by pairing a cloud-hosted large language model with adaptive typographic formatting. At its core, the tool streams a request to Google's Gemini API to generate accurate, context-aware summaries in real time, removing the need to host or fine-tune a local model. The application accepts pasted text as well as uploaded .txt, .pdf, and .docx files, extracting and summarizing each source independently. To further optimize readability, the system layers a custom half-word bolding technique with a second pass that identifies and highlights genuine keywords using part-of-speech tagging and frequency ranking, and lets users manually adjust line, word, letter, and text spacing. Deployed through a lightweight Flask web framework that proxies requests to Gemini, the system provides a highly personalized and accessible user interface. Initial results demonstrate that this integrated approach significantly improves both reading speed and comprehension, enabling targeted students to digest complex textual data with greater focus.

arXiv

ClassVision: AI-Powered Classroom Attendance System

Ankit Kumar Aggarwal, Veerabhadra Rao Marellapudi, Ovadia Sutton, Youshan Zhang

Students and working professionals have to go through the attendance process every day. Traditional methods of marking attendance using pen and paper or online platforms are human-intensive and time-consuming. To address the challenges in manual attendance processes, this research explores the use of face detection (FD) and face recognition (FR) technology to automate the attendance process, particularly in educational settings, and build a ClassVision course attendance system. We also propose an automated attendance system featuring a human-computer interaction (HCI) and user-friendly web interface that utilizes real-time image processing to identify and recognize students in classrooms and automatically record their attendance. We identified RetinaFace as the best face detection model, and when combined with Face Recognition for verification, it provided the most promising results with a cropped embedding of 50x50 pixels.

arXiv

Dynamic Coalition Formation and Communication Pricing in Skill-Based Agentic AI Systems

Mojtaba Eslami

Modern agentic AI systems combine multiple large language model agents with heterogeneous skills, yet most architectures either fix communication in advance or allow full broadcast. Both can be inefficient because token cost, latency, redundancy, and error propagation increase with the number of active agents and communication links. We model agent selection and communication as a cooperative game with task-conditioned net utility $U(C\mid x)=V(C\mid x)-\sum_{i\in C}c_i$, separating coalition-level costs from agent activation costs. We propose a marginal-value activation rule and greedy router, extend the model to optimize communication edges with per-edge costs, and use estimated Shapley values to predict which agents are worth contacting before and during execution. We connect the problem to submodular maximization and prove two limited guarantees: a curvature-refined bound for a monotone, cardinality-constrained special case, and a tight $1/2$-approximation, with a correction for signed objectives, for an unconstrained non-monotone case via double greedy. Neither guarantee applies directly to the main router, which remains a heuristic. We also prove a Shapley-submodularity sandwich bound linking the error of marginal-value routing to a per-agent diminishing-returns quantity. In synthetic experiments, greedy routing achieves $99.5%$ of brute-force-optimal utility while activating $1.96$ of $8$ agents on average, compared with $38.8%$ for full broadcast. Performance is robust to activation cost and redundancy weight but falls to $66%$ under strong violations of submodularity or noisy value estimates. We distinguish the framework from Shapley pricing, hedonic coalition formation, and communication-graph pruning, and propose evaluation on real multi-agent LLM benchmarks.

arXiv

Multimodal Wearable-Based Olfactory-Induced Emotion Recognition in Arousal-Valence Dimensions

Chen-Yang Xu, Lan Zhang, Fei-Yi Fan, Bin Hu, Qing-Hao Meng

Olfaction is important for emotion regulation because it acts as a non-intrusive and cognitively lightweight pathway that directly engages the brain s affective circuitry and achieves unobtrusive emotional modulation. This trait is essential for advancing practical affective computing in daily and attention-critical scenarios. However, current olfactory emotion research has two key limitations. First, it overemphasises the valence dimension while neglecting arousal. Second, it lacks multimodal datasets that synchronously capture central and peripheral physiological responses to olfactory stimuli. To address these issues, we construct a large-scale multimodal olfactory emotion dataset based on 111 subjects, in which odors are labeled in the 2D arousal-valence space and electroencephalogram (EEG), electrocardiogram (ECG), and photoplethysmography (PPG) signals synchronously recorded. Nevertheless, multimodal signals present challenges such as non-stationarity, differences in latency, and cross-modal heterogeneity. Thus, we propose a spatiotemporal-frequency hybrid fusion network (STF-HFNet), which integrates three core modules. Frequency aggregation processing learns adaptive frequency aggregation in order to model non-stationary dynamics. Reciprocal guided attention enables reciprocal bidirectional calibration for cross-modal temporal alignment without synchronisation priors. Hybrid collaborative fusion combines spatial and channel attention mechanisms to enhance cross-modal complementarity while suppressing redundant information. Extensive experiments show that STF-HFNet achieves state-of-the-art (SOTA) recognition accuracies of 88.34% on the AMIGOS dataset and 92.40% on our self-constructed dataset, and outperform the SOTA methods by 8.27% and 5.07%, respectively.

arXiv

HiKV: Hierarchical Importance-Aware KV Cache with Hardware Acceleration for LLM Decoding

Chao Fang, Jun Yin, Man Shi, Marian Verhelst

With the rapid adoption of long-context large language models (LLMs), the continuously growing KV cache during decoding has become the critical memory bottleneck. To tackle this challenge, we propose HiKV, a novel algorithm-hardware co-design that exploits KV cache redundancy through hierarchical importance awareness. Algorithmically, HiKV compresses the KV cache at two granularities: Stage I evicts unimportant tokens within a fixed budget, and Stage II further loads only the significant elements of each retained token, reaching compression ratios unattainable at a single granularity. Architecturally, we develop a dedicated accelerator centered on a reconfigurable importance sorter that switches between the distinct sorting datapaths each stage requires, unifying the two-stage acceleration in one circuit with minimal overhead. Evaluated on representative LLMs, HiKV achieves up to 7.95x speedup and 90% energy reduction in the attention computation over the vanilla KV cache baseline within negligible 1% accuracy loss. Under iso-accuracy constraints, HiKV outperforms state-of-the-art importance-based methods by achieving an additional 1.82~4.87x reduction in external memory accesses. These benefits are enabled by specialized hardware components that add only 8% to the system area.

Keep browsing

More papers from top AI conferences

30+ top conferences, by conference, year, and topic.