Amazon Science

@amazon.science

The latest news and research from Amazon's science community. http://www.amazon.science

Training a graph neural network on multiple objectives usually means blending conflicting gradients at every step. Instead of compromising among parameter updates from different training objectives, ControlG allocates capacity to objectives sequentially and dynamically via PID control. #ICML2026

How controllers from industrial machinery can coordinate multitask machine learning

Instead of compromising among parameter updates dictated by different training objectives, ControlG allocates computational capacity to objectives sequentially and dynamically.

amazon.science

🎉 The Chronos family of models has reached 1 billion downloads on Hugging Face: https://amzn.to/3YBkfZl Chronos-2 handles univariate, multivariate, and covariate-informed forecasting in a zero-shot manner, outperforming existing time series foundation models by a substantial margin.

Introducing Chronos-2: From univariate to universal forecasting

In-context learning enables a model that can solve forecasting tasks with an arbitrary number of dimensions in a zero-shot manner.

amazon.science

The biggest AI breakthroughs are still ahead of us. At @vivatech.bsky.social, Peter DeSantis, Amazon’s head of AI, chips, and quantum, said transformers won’t be the last architecture, and that chips and models have to evolve in lockstep to get there.

Amazon's AI SVP on why AI's biggest breakthroughs haven't happened yet

At VivaTech 2026 in Paris, Peter DeSantis argued we're at the starting line of AI development with the most important breakthroughs still ahead.

aboutamazon.com

Amazon researchers release Simple Strands Agent (SSA), a lightweight, open-source harness that achieves SOTA across SWE-Bench-Verified, SWE-Bench-Pro, and Terminal-Bench2 by closing the gap between what models intend and what harnesses execute.

Bridging intent and execution in agentic systems

The harnesses that mediate between models and tools in agentic systems are becoming their own performance bottleneck, but a few simple design principles can fix what ails them.

amazon.science

Today is CVPR Denver with AWS/Amazon. And it's all about Computer Vision. Something I haven't spent a lot of time with but decided to see how I could apply #serverless. Here is what I came up with. Check it out & if you're at CVPR come see me at the Amazon booth after 2! edjgeek.com/blog/9-tiles...

From 9 Tiles to 900: Scaling Computer Vision Pipelines — edjgeek

Computer vision pipelines are scale problems disguised as orchestration problems. You can solve both with a single function.

edjgeek.com