APIUp to 25% cheaper than official pricesTry the API →
HermesHermes Agent Docs
API
Back to News
paperPost TrainingReasoning

A Horizon Loss Beats Cross-Entropy on ImageNet Across Three Backbones

A new loss function reframes training as a planning problem, beating cross-entropy on ImageNet with a one-line change that scales with label noise.

A Horizon Loss Beats Cross-Entropy on ImageNet Across Three Backbones
Source
Google DeepMind
Published
Author
AlphaSignal Newsroom
Read
1 min read

A new loss function reframes training as a planning problem, beating cross-entropy on ImageNet with a one-line change that scales with label noise.

Reporting is indexed from AlphaSignal. Rights remain with the original publisher and cited sources.

Read original report

Next reads

Princeton's Queen Teaches a Grandmaster Chess Engine to Explain Its Moves
AlphaSignal · paper

Princeton's Queen Teaches a Grandmaster Chess Engine to Explain Its Moves

Harvard's Projection Sampling Makes Standard Fine-Tuning Match Reinforcement Learning
AlphaSignal · paper

Harvard's Projection Sampling Makes Standard Fine-Tuning Match Reinforcement Learning

AC2 Beats GRPO on Math Proofs Using 2.5x Fewer Decoding FLOPs
AlphaSignal · paper

AC2 Beats GRPO on Math Proofs Using 2.5x Fewer Decoding FLOPs