Our analysis involved a two-phase process.

We study the problem of private online learning, focusing on online prediction from experts (OPE) and online convex optimization (OCO).

Reinforcement Learning (RL) agents often learn policies that do not generalise across tasks in which the environmental features and optimal skills are different [des Combes et al., 2018, Garcin et al., 2024].

'We excel at every phase of AI': Nvidia CEO quells Wall Street fears of AI bubble amid market selloff Global share markets rose after Nvidia posted third-quarter earnings that beat Wall Street estimates, assuaging for now concerns about whether the high-flying valuations of AI firms had peaked. On Wednesday, all eyes were on Nvidia, the bellwether for the AI industry and the most valuable publicly traded company in the world, with analysts and investors hoping the chipmaker's third-quarter earnings would dampen fears that a bubble was forming in the sector. Jensen Huang, founder and CEO of Nvidia, opened the earnings call with an attempt to dispel those concerns, saying that there was a major transformation happening in AI, and Nvidia was foundational to that transformation. "There's been a lot of talk about an AI bubble," said Huang. "From our vantage point, we see something very different. As a reminder, Nvidia is unlike any other accelerator. We excel at every phase of AI from pre-training to post-training to inference."

This paper questions the effectiveness of a modern predictive uncertainty quantification approach, called evidential deep learning (EDL), in which a single neural network model is trained to learn a meta distribution over the predictive distribution by minimizing a specific objective function.

Typically, existing models are trained and tested on the same dataset, leading to poor generalizability across variations in device location, device mounting orientation, and human activity type.

Identifying the features learned by neural networks is a core challenge in mechanistic interpretability. Sparse autoencoders (SAEs), which learn a sparse, overcomplete dictionary that reconstructs a network's internal activations, have been used to identify these features. However, SAEs may learn more about the structure of the dataset than the computational structure of the network. There is therefore only indirect reason to believe that the directions found in these dictionaries are functionally important to the network. We propose end-to-end (e2e) sparse dictionary learning, a method for training SAEs that ensures the features learned are functionally important by minimizing the KL divergence between the output distributions of the original model and the model with SAE activations inserted.

A drone explodes during a Russian drone strike in Kyiv on Nov. 14. LONDON - Explosive weapons killed or injured children at record levels last year, as wars increasingly move into urban areas, Save the Children said in a report published Thursday. Nearly 12,000 children were killed or injured in conflict last year worldwide, said the U.K.-based charity, citing U.N. figures. This is the highest number since records began in 2006, and is 42% higher than the 2020 total. Previously, children in war zones were more likely to die from malnutrition, disease or failing health systems. In a time of both misinformation and too much information, quality journalism is more crucial than ever.