Researchers are leveraging neural networks to rediscover and potentially improve upon classic algorithms for multiplying matrices, pushing the boundaries of computational efficiency.
A new framework ensures machine learning models remain indistinguishable from real data, even under attack, offering robust defenses against adversarial manipulation.
A new approach to text generation dramatically boosts performance by cleverly repurposing information from previously discarded possibilities.
A new deep learning model accurately simulates human visual attention when viewing paintings, offering insights into how we perceive art.
A new pipeline combining convolutional neural networks with traditional halo finding techniques offers a faster, more efficient way to map the distribution of dark matter in the universe.
Researchers have developed a novel framework that empowers large language models to independently explore and reason over knowledge graphs, dramatically improving their ability to answer complex questions.
![Integrating weather data into load forecasting significantly enhances accuracy and reduces prediction uncertainty, particularly in regions with high temperature variability-where improvements in both mean absolute percentage error and the range of likely outcomes are substantially greater than in more temperate areas-as evidenced by the performance of five neural network architectures transitioning to a 24-hour weather integration [latex] (W=24) [/latex].](https://arxiv.org/html/2602.21415v1/x1.png)
New research rigorously compares the performance of five neural network architectures in predicting US electricity demand.
A new approach combines the power of large language models with structured knowledge and graph analytics to build autonomous agents capable of intent-driven data exploration.
The pursuit of increasingly powerful artificial intelligence is colliding with the limitations of a growth-dependent economic system, demanding a fundamental rethinking of AI’s purpose.
![The computational cost of verifying mathematical proofs-specifically, those generated by a semiautonomous system-scales relative to the complexity of the problem, as demonstrated by the inference cost per proof being benchmarked against the solution to Erdős-1051 [latex] [/latex].](https://arxiv.org/html/2602.21201v1/FP_compute.png)
A new approach, Aletheia, delivers a faster, more scalable solution for a key challenge in tensor decomposition by autonomously tackling the FirstProof problem.