Amazon Science
@amazon.science
"One workload is power-bound. The next workload is memory-bandwidth-bound. The next is memory-bound. It's one of the most interesting hardware design problems that we've seen in ages." Amazon SVP Peter DeSantis sat down with SemiAnalysis founder Dylan Patel at #AIInfraSummit:
Inside a decade of Amazon chip design, built workload by workload
At AI Infra Summit, Amazon SVP Peter DeSantis explained how real customer workloads guided every generation of Amazon's chips, from Nitro to Trainium.
www.aboutamazon.com
September 22, 2026 at 7:06 PM
"One workload is power-bound. The next workload is memory-bandwidth-bound. The next is memory-bound. It's one of the most interesting hardware design problems that we've seen in ages." Amazon SVP Peter DeSantis sat down with SemiAnalysis founder Dylan Patel at #AIInfraSummit:
Amazon Bio Discovery developed three AI approaches to accelerate antibody drug design: MochiBind (sequence-based affinity ranking), CA-MAP (developability prediction with batch effect correction), and an agent-guided design system with 46 lab-validated hits against a novel cancer target.
Advancing AI for biology: Teaching models to design and characterize antibodies
Three new papers from Amazon Bio Discovery address bottlenecks in AI-driven antibody engineering, from benchmarking binding predictors to experimentally validating de novo design.
www.amazon.science
September 21, 2026 at 7:27 PM
Amazon Bio Discovery developed three AI approaches to accelerate antibody drug design: MochiBind (sequence-based affinity ranking), CA-MAP (developability prediction with batch effect correction), and an agent-guided design system with 46 lab-validated hits against a novel cancer target.
Years of iterating against the same benchmarks should, by textbook logic, produce overfitting. It largely doesn't.
New research explains why: strategies that generalize can be expressed in too compact a form to allow memorization, while the ones that overfit don't survive a compression.
New research explains why: strategies that generalize can be expressed in too compact a form to allow memorization, while the ones that overfit don't survive a compression.
Why don’t machine learning research agents overfit?
New research indicates that AI agents learn compressible models of data, which don’t have enough space to enable memorization.
www.amazon.science
September 10, 2026 at 8:01 PM
Years of iterating against the same benchmarks should, by textbook logic, produce overfitting. It largely doesn't.
New research explains why: strategies that generalize can be expressed in too compact a form to allow memorization, while the ones that overfit don't survive a compression.
New research explains why: strategies that generalize can be expressed in too compact a form to allow memorization, while the ones that overfit don't survive a compression.
Reposted by Amazon Science
A blog post on some neat work with @zstevenwu.bsky.social and Martin Bertran: www.amazon.science/blog/why-don...
Why don’t machine learning research agents overfit?
New research indicates that AI agents learn compressible models of data, which don’t have enough space to enable memorization.
www.amazon.science
September 10, 2026 at 3:50 PM
A blog post on some neat work with @zstevenwu.bsky.social and Martin Bertran: www.amazon.science/blog/why-don...
When LLM judges agree, the right question is why. Shared prompts, model families, or training lineage can make a majority look stronger than it is. Dependence-aware aggregation via Ising models accounts for this, improving accuracy 9–14% over weighted majority vote.
When LLM judges agree, should we believe them?
Discounting the opinions of LLM judges with highly correlated outputs ensures that panels of judges reflect a true diversity of perspectives.
www.amazon.science
August 28, 2026 at 6:29 PM
When LLM judges agree, the right question is why. Shared prompts, model families, or training lineage can make a majority look stronger than it is. Dependence-aware aggregation via Ising models accounts for this, improving accuracy 9–14% over weighted majority vote.
How did a model upgrade make agents worse? By pairing real enterprise SOPs with functioning tools and ground-truth grading across 12 industries and 2,000+ tasks, SOP-Bench helps find such anomalies.
SOP-Bench: A new benchmark for evaluating AI agents on real business procedures
Extendable framework enables testing agents on the full set of capabilities required to successfully complete a procedure, not isolated proxy tasks.
www.amazon.science
August 25, 2026 at 3:59 PM
How did a model upgrade make agents worse? By pairing real enterprise SOPs with functioning tools and ground-truth grading across 12 industries and 2,000+ tasks, SOP-Bench helps find such anomalies.
Amazon's Automated Reasoning Group started by demoing tools to prove AWS systems secure and correct.
A decade later, they have proved the Nitro Isolation Engine, cryptographic code, and S3 correct. Now they are applying the same techniques to AI.
A decade later, they have proved the Nitro Isolation Engine, cryptographic code, and S3 correct. Now they are applying the same techniques to AI.
A decade of mathematical certainty: Reflections on the Automated Reasoning Group
Ten years after we founded the Automated Reasoning Group, mathematical logic has moved from academic research into production services that secure millions of customer workloads — demonstrating that s...
www.amazon.science
August 11, 2026 at 4:33 PM
Amazon's Automated Reasoning Group started by demoing tools to prove AWS systems secure and correct.
A decade later, they have proved the Nitro Isolation Engine, cryptographic code, and S3 correct. Now they are applying the same techniques to AI.
A decade later, they have proved the Nitro Isolation Engine, cryptographic code, and S3 correct. Now they are applying the same techniques to AI.
Reposted by Amazon Science
🎉 Congratulations to A/Prof. Wei Bao, one of the 34 recipients of the Amazon "Build on Trainium" Research Award!
Wei will explore the Trainium architecture for "FACTOR: Federated Adversarial Co-Training with Textual Gradient for LLM Security and Robustness."
www.amazon.science/research-awa...
Wei will explore the Trainium architecture for "FACTOR: Federated Adversarial Co-Training with Textual Gradient for LLM Security and Robustness."
www.amazon.science/research-awa...
34 Amazon Research Awards Build on Trainium recipients announced
Amazon announces 34 recipients of the Build on Trainium program, a $110 million credit initiative supporting AI research at 30 universities including Stanford, UC Berkeley, UIUC, UCLA, CMU, and MIT, w...
www.amazon.science
August 10, 2026 at 12:05 AM
🎉 Congratulations to A/Prof. Wei Bao, one of the 34 recipients of the Amazon "Build on Trainium" Research Award!
Wei will explore the Trainium architecture for "FACTOR: Federated Adversarial Co-Training with Textual Gradient for LLM Security and Robustness."
www.amazon.science/research-awa...
Wei will explore the Trainium architecture for "FACTOR: Federated Adversarial Co-Training with Textual Gradient for LLM Security and Robustness."
www.amazon.science/research-awa...
Attending #KDD2026? Come meet Amazonians and fellow attendees at our mixer. Spots are limited – request to attend:
Amazon After Hours @ KDD · Luma
Join us for a relaxed evening away from the conference buzz, where you can connect with Amazonians and fellow KDD attendees in a fun, casual setting. If you...
luma.com
August 6, 2026 at 9:59 PM
Attending #KDD2026? Come meet Amazonians and fellow attendees at our mixer. Spots are limited – request to attend:
📣 Amazon Research Awards announces the 34 recipients of the Build on Trainium program, a $110M credit initiative supporting AI research at 30 universities on AWS Trainium:
34 Amazon Research Awards Build on Trainium recipients announced
Amazon announces 34 recipients of the Build on Trainium program, a $110 million credit initiative supporting AI research at 30 universities including Stanford, UC Berkeley, UIUC, UCLA, CMU, and MIT, w...
www.amazon.science
August 5, 2026 at 3:05 PM
📣 Amazon Research Awards announces the 34 recipients of the Build on Trainium program, a $110M credit initiative supporting AI research at 30 universities on AWS Trainium:
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.
www.amazon.science
July 31, 2026 at 7:18 PM
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
As AI agents take on higher-stakes decisions, their actions need to be provably correct. Amazon is investing in the Lean FRO to make mathematical proof accessible to every developer: amzn.to/4vUsCg9
July 27, 2026 at 7:08 PM
As AI agents take on higher-stakes decisions, their actions need to be provably correct. Amazon is investing in the Lean FRO to make mathematical proof accessible to every developer: amzn.to/4vUsCg9
Turnstile is an open-source Rust proxy that records token-native rollout data for agentic RL. Harness stays unchanged. Trainer gets exact token IDs, log probs, and loss masks.
Capturing token IDs during agentic interactions for better reinforcement learning
A new Rust proxy called Turnstile sits between the model backend and the agent harness to capture information lost in mere text transcripts.
www.amazon.science
July 20, 2026 at 4:58 PM
Turnstile is an open-source Rust proxy that records token-native rollout data for agentic RL. Harness stays unchanged. Trainer gets exact token IDs, log probs, and loss masks.
🎉 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.
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.
www.amazon.science
July 17, 2026 at 8:25 PM
🎉 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.
Chronos-2 handles univariate, multivariate, and covariate-informed forecasting in a zero-shot manner, outperforming existing time series foundation models by a substantial margin.
HydroShear is a new physics-based simulator that teaches robots how to use their sense of touch to perform complex manipulation tasks, in a way that transfers seamlessly to the real world:
Amazon and University of Michigan give robots a sense of touch
HydroShear, a new physics-based simulator, teaches robots how to use their sense of touch to perform complex manipulation tasks, in a way that transfers seamlessly to the real world.
www.amazon.science
July 15, 2026 at 9:54 PM
HydroShear is a new physics-based simulator that teaches robots how to use their sense of touch to perform complex manipulation tasks, in a way that transfers seamlessly to the real world:
What a week. #ICML2026 brought together some of the best minds in machine learning, and we were proud to be part of it. Thank you to everyone who joined us in Seoul. See you next year!
July 13, 2026 at 5:36 PM
What a week. #ICML2026 brought together some of the best minds in machine learning, and we were proud to be part of it. Thank you to everyone who joined us in Seoul. See you next year!
If you haven't stopped by our booth yet, come find us before the #ICML2026 exhibit wraps up — more research talks and opportunities to connect with our scientists today. Full schedule: https://amzn.to/3SQwSPZ
July 8, 2026 at 11:45 PM
If you haven't stopped by our booth yet, come find us before the #ICML2026 exhibit wraps up — more research talks and opportunities to connect with our scientists today. Full schedule: https://amzn.to/3SQwSPZ
Amazon researchers have accepted papers at #ICML2026 spanning machine learning, causal reasoning, LLM inference, agentic systems, vision-language models, graph learning, robotics, and more. Explore the full list:
ICML 2026
The International Conference on Machine Learning (ICML) is the premier gathering of professionals dedicated to the advancement of the branch of artificial intelligence known as machine learning.
www.amazon.science
July 8, 2026 at 4:12 AM
Amazon researchers have accepted papers at #ICML2026 spanning machine learning, causal reasoning, LLM inference, agentic systems, vision-language models, graph learning, robotics, and more. Explore the full list:
Another great day ahead at #ICML2026. Stop by our booth for research talks, a chance to meet our scientists, and to explore opportunities at Amazon. Full schedule: https://amzn.to/4f3pSGP
July 7, 2026 at 11:20 PM
Another great day ahead at #ICML2026. Stop by our booth for research talks, a chance to meet our scientists, and to explore opportunities at Amazon. Full schedule: https://amzn.to/4f3pSGP
We're looking forward to connecting with the machine learning community at #ICML! We have talks happening at our booth all week, plus a chance to connect 1:1 with our researchers.
Full schedule: https://amzn.to/3SBXPXA #ICML2026
Full schedule: https://amzn.to/3SBXPXA #ICML2026
July 7, 2026 at 1:33 PM
We're looking forward to connecting with the machine learning community at #ICML! We have talks happening at our booth all week, plus a chance to connect 1:1 with our researchers.
Full schedule: https://amzn.to/3SBXPXA #ICML2026
Full schedule: https://amzn.to/3SBXPXA #ICML2026
We're open at @icmlconf.bsky.social! Find us at booth B207 for live demos, time with our scientists, and a chance to learn about opportunities at Amazon. #ICML2026
July 7, 2026 at 4:00 AM
We're open at @icmlconf.bsky.social! Find us at booth B207 for live demos, time with our scientists, and a chance to learn about opportunities at Amazon. #ICML2026
Attending #ICML2026? Our booth is packed with talks and opportunities to connect with our scientists, covering agentic AI, LLM training, reinforcement learning, and more.
Check out the full schedule: https://amzn.to/4p2LgAy
Check out the full schedule: https://amzn.to/4p2LgAy
July 6, 2026 at 10:08 AM
Attending #ICML2026? Our booth is packed with talks and opportunities to connect with our scientists, covering agentic AI, LLM training, reinforcement learning, and more.
Check out the full schedule: https://amzn.to/4p2LgAy
Check out the full schedule: https://amzn.to/4p2LgAy
Amazon researchers have accepted papers at @icmlconf.bsky.social spanning machine learning, causal reasoning, LLM inference, agentic systems, vision-language models, graph learning, robotics, and more. Explore the full list. #ICML2026
ICML 2026
The International Conference on Machine Learning (ICML) is the premier gathering of professionals dedicated to the advancement of the branch of artificial intelligence known as machine learning.
www.amazon.science
July 5, 2026 at 9:54 AM
Amazon researchers have accepted papers at @icmlconf.bsky.social spanning machine learning, causal reasoning, LLM inference, agentic systems, vision-language models, graph learning, robotics, and more. Explore the full list. #ICML2026
Amazon's latest machine learning research is headed to Seoul. We'll be at ICML with accepted papers, live demos, and researchers presenting across agentic AI, robotics, and more: https://amzn.to/4vHY5CJ #ICML2026
July 2, 2026 at 11:30 PM
Amazon's latest machine learning research is headed to Seoul. We'll be at ICML with accepted papers, live demos, and researchers presenting across agentic AI, robotics, and more: https://amzn.to/4vHY5CJ #ICML2026