Quantization, distillation, pruning and serving work aimed at the same accuracy for less memory, latency and money.
arXiv:2609.01212v1 Announce Type: new Abstract: With the rapid and continuous growth in the incorporation of machine learning models based on the Transformer architecture, capable deployment is in hi...
arXiv:2609.00591v1 Announce Type: new Abstract: An image may be worth a thousand words, but most captioning models describe it in only a few. Modern vision-language models produce fluent high-level c...
The paper introduces Inverted Contrastive Learning for Unsupervised Feature Selection (ICLFS), a method that treats each feature as a sample by inverting the data matrix and applies a contrastive learning framework to learn consistent representations across masked positive views and a shuffled negative view. Feature saliency is derived from the magnitude of projector‑space embeddings, and a Laplacian‑Gated Ranking Correction step refines the ranking by reducing local redundancy. Experiments on 12 benchmark datasets show that ICLFS achieves the best clustering accuracy on 10 datasets compared to both classical and neural baselines, demonstrating the effectiveness of feature‑wise contrastive consistency for unsupervised feature selection.
The paper introduces STORM, a one‑stage generative AI framework that reformulates Earth system data assimilation as diffusion‑based Bayesian posterior sampling, replacing costly PDE ensemble forecasts with scalable AI inference. STORM employs a spatiotemporal transformer with a global‑attention algorithm that reduces computational complexity from quadratic to linear, enabling high‑resolution, long‑context modeling. The system scales to 74,400 GPUs on Frontier, achieving 96–99 % strong‑scaling efficiency and up to 6 ExaFLOPs sustained BF16 throughput, while supporting 32,768‑member ensembles for uncertainty quantification in just 34 seconds on 4,096 GPUs, and demonstrates improved hurricane tracking and climate reanalysis accuracy.
The paper introduces Iterative MBR Distillation for Error Span Detection (ESD) in machine translation, a self‑evolution framework that replaces human annotations with pseudo‑labels generated by a large language model. By iteratively applying Minimum Bayes Risk decoding, the method produces high‑quality error spans without costly human effort. Experiments on WMT Metrics Shared Task datasets show that models trained solely on these pseudo‑labels outperform both unadapted baselines and supervised models trained on human data at system and span levels, while keeping sentence‑level performance competitive.
The paper evaluates the effectiveness of tolerance‑based conformance tests for INT8 quantized GEMM kernels used in large language models. By injecting nine faults into a Qwen3‑1.7B reference pipeline, the authors show that most faults shift outputs by at most one bfloat16 spacing, rendering a tolerance of one spacing blind to these errors. They further demonstrate that requantizing weight scales to the nearest power of two aligns CUTLASS and Triton implementations bit‑for‑bit and produces identical token sequences, with only minor perplexity changes.
Hybrid event-frame sensors combine an Event Vision Sensor (EVS) and an Active Pixel Sensor (APS) on a single chip, offering high dynamic range, low latency, and rich spatial intensity data. The paper introduces a unified statistics-based noise model that captures photon shot noise, dark current noise, fixed-pattern noise, and quantization noise for both APS and EVS pixels, and links EVS noise to illumination and dark current. It also presents a calibration pipeline to estimate these noise parameters from real data and proposes H-ESIM, a simulator that generates realistic RAW frames and events, validated on two hybrid sensors for tasks such as video frame interpolation and deblurring.
arXiv:2609.00654v1 Announce Type: new Abstract: We describe the SciTrue team's participation in both subtasks of the NTCIR-19 SciClaimEval task~\cite{sciclaimeval}, which asks systems to verify scien...
The paper introduces Faster Flash Decoding (FFD), a hardware‑algorithm co‑design that fuses the selector and computation into a single kernel to eliminate memory‑bandwidth bottlenecks in long‑context decoding. By replacing metadata indices with low‑bit quantized, content‑aware scanning and employing a top‑delta strategy for dynamic block filtering, FFD achieves up to 11.6× kernel‑level speedup and scales to 256K context length while preserving model accuracy. The approach is training‑free, plug‑and‑play, and demonstrates significant throughput gains on benchmarks such as RULER and LongBench.
The paper investigates subliminal learning, where hidden traits from a teacher model are transferred to a student during distillation. It introduces trait‑direction drift as the underlying mechanism, showing that biased generation creates measurable preference gaps that accumulate into behavioral transfer during fine‑tuning. The authors propose probe‑space corridor regularization, a targeted defense that constrains drift along a calibrated trait direction, significantly reducing hidden‑trait transfer while maintaining task performance.
SinkPruner is a training‑free framework that prunes visual tokens for multimodal large language models by first removing high‑norm redundant tokens with a visual sanitizer and then selectively keeping tokens that align with the text query using a text‑guided pruner. The coarse‑to‑fine design reduces attention sink and dispersion, enabling an 89% token reduction while preserving 96.5% of LLaVA‑1.5’s performance and 91.8% of Qwen2.5‑VL’s performance across twelve image‑language and four video‑language benchmarks. The visual sanitizer also improves existing pruning methods, showing strong transferability.
Large Language Models suffer from high inference latency, especially when generating long chains of thought. Existing speculative decoding methods draft and verify tokens in parallel but ignore semantic equivalence, causing inefficient rejections. The proposed SemanticSpec framework verifies entire semantic sequences by probing internal hidden states, achieving up to 2.7× speedup on DeepSeekR1-32B and 2.1× on QwQ-32B while outperforming token‑level and sequence‑level baselines in both efficiency and effectiveness.
Long-context inference retains a growing key--value (KV) cache during decoding, which consumes substantial GPU memory and can reduce generation throughput. This bottleneck remains in hybrid language m...
Enterprise wireless access points (APs) are promising platforms for predictive machine learning (ML), but their primary responsibility remains providing wireless connectivity and network services. Pre...
Compact instruction-following rerankers are attractive for deployment, but conventional distillation pipelines typically train students by offline imitation of teacher outputs on a fixed set of exampl...
Vision Transformers (ViTs) are increasingly used in split-inference systems, where edge devices transmit intermediate token representations to a remote cloud. In this setting, token reduction lowers c...
Satellites are emerging as promising platforms to extend generative \emph{artificial intelligence} (AI) services to remote areas lacking terrestrial infrastructure. However, deploying space generative...
Mixture-of-Experts (MoE) architectures scale Large Language Model (LLM) capacity efficiently by activating a sparse subset of experts per token. However, modern MoE inference remains heavily constrain...
arXiv:2604.26582v2 Announce Type: replace-cross Abstract: Reliable celestial attitude determination is a critical requirement for autonomous spacecraft navigation, yet traditional "Lost-in-Space" (LI...
arXiv:2608.29263v1 Announce Type: new Abstract: Large Language Models (LLMs) often suffer from hallucination and struggle with complex reasoning tasks requiring multi-hop domain knowledge. While inte...