Squashing 'fantastic bugs' hidden in AI benchmarks
After reviewing thousands of benchmarks used in AI development, a Stanford team found that 5% could have serious flaws with far-reaching ramifications.
Dec 11, 2025
0
40
Computer Sciences
Page 23: Research news on Machine learning methodologies
Machine learning methodologies encompass algorithmic frameworks and architectures for training, optimizing, and deploying models such as neural networks, transformers, diffusion models, and reinforcement learning agents. Work in this area develops new training objectives, curriculum schemes, speculative and efficient decoding, pruning and communication-reduction strategies, and biologically inspired or physics-informed architectures. The domain also includes safety preservation, unlearning, scaling laws, and specialized methods for vision, language, control, and scientific computing, aiming to improve performance, efficiency, robustness, and controllability of complex AI systems.
Computer Sciences
AI training method slashes pre-training time by 50% while boosting accuracy
A much faster, more efficient training method developed at the University of Waterloo could help put powerful artificial intelligence (AI) tools in the hands of many more people by reducing the cost and environmental impact ...
Dec 9, 2025
0
60
Computer Sciences
Overparameterized neural networks: Feature learning precedes overfitting, research finds
Modern neural networks, with billions of parameters, are so overparameterized that they can "overfit" even random, structureless data. Yet when trained on datasets with structure, they learn the underlying features.
Dec 5, 2025
0
56
Computer Sciences
Making simulations more accurate than ever with deep learning
Future events such as the weather or satellite trajectories are computed in tiny time steps, so the computation must be both efficient and as accurate as possible at each step lest errors pile up. A Kobe University team has ...
Dec 4, 2025
0
29
Computer Sciences
Algorithm offers faster, more reliable control over language model outputs
A paper co-authored by Prof. Alex Lew has been selected as one of four "Outstanding Papers" at this year's Conference on Language Modeling (COLM 2025), held in Montreal in October.
Dec 4, 2025
0
39
Hi Tech & Innovation
Biological intelligence as the basis for new AI systems
In a new research project led by the Central Institute of Mental Health (CIMH) in Mannheim, scientists are investigating how insights into learning processes in animal brains can be used to make artificial intelligence (AI) ...
Dec 4, 2025
0
31
Business
Platforms like TikTok could boost viewing times by grouping users to better match their preferences
In the nine years since TikTok debuted, it's helped transform the way people view and absorb information, along with other short-form video platforms such as Instagram and Snapchat. Every month, TikTok alone has nearly 1.6 ...
Dec 4, 2025
0
19
Computer Sciences
Guided learning helps previously 'untrainable' neural networks learn effectively
Even networks long considered "untrainable" can learn effectively with a bit of a helping hand. Researchers at MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) have shown that a brief period of alignment ...
Dec 3, 2025
0
26
Machine learning & AI
Taming chaos in neural networks: A biologically plausible way
A new framework that causes artificial neural networks to mimic how real neural networks operate in the brain has been developed by a RIKEN neuroscientist and his collaborator. In addition to shedding light on how the brain ...
Dec 3, 2025
0
97
Computer Sciences
Approximate domain unlearning: Enabling safer and more controllable vision-language models
Vision-language model (VLM) is a core technology of modern artificial intelligence (AI), and it can be used to represent different forms of expression or learning, such as photographs, illustrations, and sketches.
Dec 2, 2025
0
19
