Quantization, distillation, pruning and serving work aimed at the same accuracy for less memory, latency and money.
arXiv:2608.22704v1 Announce Type: new Abstract: Long-form audio inputs make the KV cache the dominant memory cost of speech LLMs. Prefill-only KV compression methods permanently discard audio KV posi...
arXiv:2608.21764v1 Announce Type: cross Abstract: Event-based vision has emerged as a promising paradigm for energy-aware artificial intelligence (AI), offering sparse, low-latency visual signals tha...
The paper introduces a training‑free structured suffix modeling technique to accelerate Diffusion Language Models (DLMs). It partitions the suffix into local, middle, and tail regions, retaining varying numbers of tokens per region and incorporating previous decoding results into current token representations. Experiments on three DLMs show significant speedups—up to 72.81× in long‑sequence inference—while often improving performance, and the method is compatible with existing acceleration strategies.
arXiv:2606.14782v3 Announce Type: replace-cross Abstract: Multimodal Large Language Models (MLLMs) achieve strong vision-language reasoning but incur large KV caches and high decoding latency with lo...
arXiv:2608.22364v1 Announce Type: new Abstract: World action models (WAMs) couple visual future prediction with robot action generation, but accelerated students can lose task capabilities during dis...
arXiv:2608.22196v1 Announce Type: cross Abstract: While cascaded multi-talker ASR (MT-ASR) leverages state-of-the-art foundation models, its performance is often capped by speaker leakage during sepa...
Interpretable AI with Local Distillation proposes a method where a black‑box teacher model guides a regularized linear student model at each query point. The teacher defines locality by upweighting training observations with similar predicted outcomes and anchors the fit with its own prediction at the query point, treated as a pseudo‑observation. By adding Gaussian randomization and refitting, the approach identifies reliable features and stable subgroups, achieving near‑teacher accuracy while producing sparse, locally interpretable linear models.
DIAG is a Diagnostic Iterative Alignment and Generation framework designed to improve data efficiency in aligning large language models for mathematical reasoning. It adaptively reshapes the practice distribution by first diagnosing valid preference-pair yield to calibrate exploration and exploitation, then generating targeted practice from the model’s failure traces. The approach is theoretically framed as a teacher‑mediated approximation to KL‑regularized reweighting, and experiments show that DIAG increases preference-pair yield and reasoning performance under the same training budget.
arXiv:2608.22963v1 Announce Type: new Abstract: Multimodal Large Language Models (MLLMs) are increasingly deployed as multi-step agents, where explicit reasoning supports task decomposition and tool...
arXiv:2608.22753v1 Announce Type: new Abstract: Large language models (LLMs) excel at text understanding and generation, yet still struggle to reliably understand and apply externally provided proced...
The paper evaluates how well Vision Transformers (ViTs) can handle token merging techniques—specifically ToMe and Mutual Pair Merging—across wheat phenotyping tasks such as growth-stage classification, wheat-head detection, and wheat-organ segmentation. It benchmarks task quality, throughput, token count, and GPU memory usage, including tests on a Raspberry Pi 5. Results show that classification is highly tolerant to token merging, whereas detection and segmentation suffer due to factors like repeated instances, thin organs, dense boundaries, and runtime overhead, and that optimized attention backends can negate apparent speed gains.
arXiv:2608.21362v1 Announce Type: new Abstract: Transformer-based large language models (LLMs) incur high prefill latency because key-value (KV) tensors must be recomputed for each request. Existing...
arXiv:2607.12550v3 Announce Type: replace-cross Abstract: The key-value (KV) cache has become the dominant memory cost of transformer inference: it grows with batch size, context length, and depth, a...
arXiv:2608.23391v1 Announce Type: cross Abstract: Structured data exists in many forms (tables, knowledge graphs, charts, and time series), and converting it into text may involve different generatio...
FedCC is a new algorithm for distillation-based federated learning that tackles label distribution skew by allowing clients to mark ambiguous samples as 'unknown' instead of forcing a potentially wrong classification. By adding this extra class and calibrating pseudo-labels on a public dataset, FedCC balances confidence across majority and minority classes. Experiments show that FedCC outperforms existing methods, achieving 67.3% accuracy even when each client has data from only one of ten classes, whereas baselines drop to near-random performance.
arXiv:2608.22926v1 Announce Type: new Abstract: Gaze is increasingly used as an input signal for vision and multimodal models, yet no consensus exists on how to represent it across datasets. Raw trac...
arXiv:2608.21811v1 Announce Type: new Abstract: Reinforcement learning for vision-language math reasoning starves under sparse reward: on a pool of 20,830 visual-math problems where Qwen2-VL-2B answe...
SlideGen is a collaborative vision‑language multi‑agent framework designed to generate scientific presentation slides from research papers. It assigns specialized agents to outline the presentation structure, align figures and tables with key claims, generate speaker notes, and compose editable PPTX slides using a diverse layout library. The system introduces a geometry‑aware density metric to evaluate visual clutter and demonstrates significant improvements in layout balance, content coverage, and text coherence over existing baselines on a 200‑paper benchmark.
arXiv:2608.23318v1 Announce Type: new Abstract: Hint-based reinforcement learning addresses reward sparsity in long-horizon agentic tasks by retaining a prefix of an expert trajectory before each rol...
arXiv:2608.21874v1 Announce Type: cross Abstract: Deep learning has achieved strong performance in encrypted traffic classification (ETC), yet its computational cost limits deployment on resource-con...