A study on on-device English-to-Traditional-Chinese subtitle translation shows that reducing the vocabulary of the LMT-60-0.6B model from 151,000 to 64,000 tokens improves speed and privacy. The Local…
Adaptive Model Compression (AMC) reduces energy consumption by 59.2% and boosts throughput 2.24x on 45nm CMOS edge devices, with only a 3.6% accuracy trade-off. The saliency-driven framework dynamical…
The SCOPE-RL framework, a two-stage reinforcement learning approach for large language models, improved average accuracy by up to 11.2 percentage points and reduced reasoning tokens by 27.1% on models…
Magnitude, a mathematical concept rooted in metric spaces, is being applied to time series analysis to improve machine learning model performance. Researchers have derived new invariants for periodic …
Two new methods using generative networks are enhancing the speed and precision of Large Hadron Collider simulations. The first method, averaging amplification, employs Bayesian networks or ensembling…
A new Likelihood Matching approach refines diffusion model training by aligning data distribution likelihoods, promising efficiency and accuracy in model generation. The method uses a quasi-likelihood…
Researchers have introduced a novel approach using normalizing flows, specifically RealNVP, to model human body poses in 6D rotation format, offering a new probabilistic foundation for motion capture.…
Researchers have developed a deep-learning method for MR elastography that reconstructs high-resolution images from highly undersampled data, achieving 2mm isotropic resolution in one minute with a to…
The PnPBO framework, a plug-and-play architecture for bilevel optimization, achieves optimal sample complexity comparable to single-level optimization by integrating modern unbiased and biased stochas…
A neural network-based model developed by researchers replicates how infants learn to control their movements in the mobile paradigm, revealing that spontaneous movements after disconnection are not r…
Diffusion generative models are accelerating AI sampling by exploiting hidden low-dimensional structures, with the denoising diffusion implicit model (DDIM) and denoising diffusion probabilistic model…
Machine learning-generated beliefs are driving a new form of statistical discrimination, according to a study that examines belief-contingent interventions like 'common identity' as potential remedies…
Researchers introduced VER (Vigilant Evaluator of Representations), a conceptual framework designed to diagnose hidden flaws in AI representations that traditional performance metrics overlook. VER de…
NVIDIA's GB300 NVL72 platform delivers up to 25x performance per watt over its predecessor, the NVIDIA Hopper, redefining AI infrastructure efficiency. The company's extreme codesign from silicon to s…
IBM's Q2 revenue fell short of expectations, but the miss reflects a strategic budget rotation from software to hardware, not weak AI demand, according to the company. Enterprises are reallocating spe…
A new study using four years of data from Belgian offshore wind farms shows that tree-based machine learning models, particularly conditional diffusion models, improve wind power forecast accuracy by …
Researchers introduced a novel regularizer based on conditional mutual information to enforce separation in deep learning models, balancing fairness and utility without sacrificing performance. The st…
A new algorithm inspired by knowledge distillation allows machine learning platforms to probe peer model predictions, escaping the overspecialization trap that limits them to current users. Semi-synth…
CoGenCast, a hybrid generative framework combining pre-trained large language models with flow-matching, achieves state-of-the-art results on multiple time series forecasting benchmarks by enabling mu…
Researchers have introduced BalDRO, a novel framework for balanced unlearning in large language models that addresses asynchronous forgetting by focusing on hard-to-unlearn data distributions. BalDRO,…