Blog editor at Amazon Science
Cambridge, MA
Joined May 2018
- Tweets146
- Following69
- Followers224
- Likes80
I'm particularly pleased about today's Amazon Science article on the Amazon editors' picks of the 20 best science books of the year (so far) as two of the selections are by my old comrades @JSEllenberg and @stevenbjohnson.
amazon.science/amazon-announ…
Larry Hardesty retweeted
On the cusp of Super Bowl LV, we take a look at how Amazon scientists are contributing to the design of the @NFL's @NextGenStats, powered by AWS.
#AmazonScience #SBLV #BigGame
amzn.to/2N2ssmX
One-shot TTS expression transfer. I'd never heard of householder flows before, but they seem pretty cool.
Last week at @ieeeICASSP, Amazon scientists showed how to adapt a text-to-speech system to a new expressive style on the strength of just one one-second example. Read more: amazon.science/blog/speech-s… #AmazonScience #IEEE #ICASSP2020
These conference-related interviews are becoming my favorite posts to do. I learn so much from them. #AmazonScience
Alexa scientist Ariya Rastrow on the blurring boundaries between acoustic processing and language understanding. Read more, and learn about Amazon’s presence at @ieeeICASSP this year: amazon.science/blog/icassp-w… #AmazonScience #ICASSP2020
"One simple trick" enables a multisource localization system to improve performance by 15%. #AmazonScience
At @ieeeICASSP, we’ll be presenting a new way to avoid the "permutation problem" that arises when an AI-powered microphone array tries to localize several sound sources at once, with a 15% improvement over the state of the art: amazon.science/blog/locating… #AmazonScience #ICASSP2020
I really enjoyed working on this article. Stefano gave me a new way to think about a lot of the AI topics that I've encountered -- and he answered my one big question about few-shot learning. #AmazonScience
On the cusp of @iclr_conf, Stefano Soatto, Amazon AWS's director of applied science for AI applications, explains how the conference's subspecialty — learning representations — came to dominate mainstream AI. amazon.science/blog/iclr-the… #AmazonScience #ICLR2020 #AI
Larry Hardesty retweeted
An @iclr_conf paper by @rasoolfa @pratikac @smolix & Stefano Soatto presents a new approach to meta-reinforcement learning that lets robots adapt to new tasks quickly, reducing the need for costly interactions with the environment amazon.science/blog/how-robo… #AmazonScience #ICLR2020
Meet Inderjit Dhillon, one of the many Amazon Fellows helping us innovate through academic collaborations.
An article on my research work at Amazon amazon.science/how-amazon-fe… #amazonscience
New method for entity alignment yields 10% increase over state of the art in area under the precision-recall curve (PRAUC). #AmazonScience
In a paper for @TheWebConf, Amazon researchers show how to merge two knowledge graphs, more accurately reconciling different names for the same entities. Read more: amazon.science/blog/combinin… #AmazonScience #TheWebConf
Cool ML approach to sharpening the results of Amazon's product discovery engine that essentially uses unlabeled data.
When a customer looks for "waterproof shoes", does she want office wear or hiking boots? In an @ACM_CHIIR paper, Amazon researchers show that inferring customers' "context of use" could improve product discovery. Read more: amazon.science/blog/why-do-y… #AmazonScience
I know we've all been reading about this for years now, but GAN-based image generation still blows my mind.
Learn about a new AI model from Amazon scientists that generates images from text, which could help refine customer discovery query results for billions of products: bit.ly/2x4jyxt #AmazonScience
Larry Hardesty retweeted
8 machine learning and conversational-AI innovations that Amazon researchers will present at this year's @RealAAAI conference, which starts tomorrow: amazon.science/blog/amazon-b… #AI #ML #AAAI20 #AmazonScience
50% reduction in SOA error rate -- that's a lot!
At @RealAAAI, @amoschitti1 and team will present a way to train an answer sentence selection system — part of a question-answering system — that yielded a 50% error rate reduction over the state of the art on a benchmark data set: amazon.science/blog/on-bench… #AAAI20 #AmazonScience
Hard to believe that four papers on one theme could display such breadth.
Our latest blog post covers 4 Amazon papers that will be presented at @WSDMSocial next month, and includes an interview with @Yoellem, VP of research & science for Alexa Shopping, who was there when the conference was established: amazon.science/blog/web-sear… #WSDM2020 #AmazonScience
Learned a lot about differential privacy from this post.
At @WSDMSocial, Amazon researchers will show how to protect privacy during large-scale analyses of textual data by re-phrasing supplied texts and basing analysis on the new phrasing. (The method uses the analytic framework of metric differential privacy.) amazon.science/blog/preservi…
This is a really beautiful paper. Important problem, innovative approach, elegant math, huge impact.
How the #AlexaAI team taught @alexa99 to correct her own defects. (Hint: using customers’ rephrased requests as implicit error signals.) amazon.science/blog/how-we-t… #AmazonScience #MachineLearning #AI #ML
Larry Hardesty retweeted
How deep-learning-based voice conversion technology enabled Alexa’s popular whisper mode to expand to all locales: amazon.science/blog/the-rese… #Alexa #AmazonScience