Operational Efficiency
Improve operational efficiency by reducing abuse, improving address quality and increasing payment/delivery success rates.
About The Event
Amazon Research Days is a global engagement program created in 2018 to connect the scientific community at Amazon, industry leaders and external academic researchers in the field of machine learning around the world. We aim to create an opportunity for leading academic and industry researchers in machine learning to share ideas and foster collaboration through both formal and informal events at the conference.
When
17th & 18th Nov, 2021
ML@Amazon
Machine Learning, Big Data and related quantitative sciences have been strategic to Amazonfrom the early years. Amazon has been a pioneer in areas such as recommendation engines, ecommerce fraud detection and large-scale optimization of fulfillment center operations.
As Amazon has rapidly grown and diversified, the opportunity for applying machine learning has exploded. We have a very broad collection of practical problems where machine learning systems can dramatically improve the customer experience, reduce cost, and drive speed and automation. These include automated pricing and demand forecasting for hundreds of millions of products, predicting ad click probabilities, ranking product search results, matching products from multiple sources, classifying products into large taxonomies, information extraction and sentiment analysis for product reviews, voice recognition, natural language understanding, question answering and conversational systems.
Below are some of the problems we’re working on
Customer Experience
Improve customer experience by increasing search results and ads relevance, personalization, and providing conversational interfaces.
Seller Experience
Improve seller experience by providing automated tools (e.g. for forecasting product demand) that help them streamline their operations and grow their businesses on Amazon.
Catalog Quality
Improve catalog quality by identifying and correcting defects related to product attributes (e.g. title) and images.
ML Platforms
Build ML platforms to speed up development and deployment of ML solutions.
ML Education
Spread ML knowledge to empower the Amazon technical community.
Talks
Welcome Keynote
Muthu Muthukrishnan
VP, Amazon Ads
Seattle
Muthu Muthukrishnan is the Vice President of Sponsored Products, a self-service advertising solution that drives product discovery and sales on Amazon.com. Muthu is a Scientist interested in Algorithms as well as Auctions and Game Theory. He is a fellow of ACM and winner of the Imre Simon Test of Time award for count-min sketch. He is excited about the combination of AI and Economics that brings shoppers and selling customers together and drives the online advertising business.
3D Visual Understanding with Neural Priors
Simon Lucey
Professor
Adelaide
Simon Lucey (Ph.D.) is a professor at the University of Adelaide where he is the Director of the Australian Institute of Machine Learning (AIML). Prior to this he was an associate professor at Carnegie Mellon University's Robotics Institute (RI). He has received various career awards including an ARC Future Fellowship (2009-2013). Simon’s research interests span computer vision, and robotics. He enjoys building computational models that underlie the processes of visual perception.
Simple and effective approaches for instance recognition
Chunhua Shen
Principal Applied Scientist, Amazon ML
Adelaide
Chunhua Shen studied at Nanjing University, at Australian National University, and received his PhD degree from the University of Adelaide. From 2012 to 2016, he held an Australian Research Council Future Fellowship. He has been working in Computer Vision for about 20 years. His worked at various places including National ICT Australia, Australian National University, Amazon and University of Adelaide. His Google scholar citation is 32000 with an H index of 90.
Spoken Language Understanding for the Indic Region
Anurag Dwarakanath
Science Manager, Amazon Alexa
Bangalore
Anurag Dwarakanath is an applied science manager in Alexa AI, building ML models for the Natural Language Understanding components of Alexa. His interests include multi-lingual natural language processing, robustness in deep learning and verification & validation of deep learning systems. Anurag holds a PhD from Indian Institute of Management Calcutta where he studied the application of Graph Theory in Wireless Sensor Networks. Anurag has over 20 publications and 15 patents.
Multilingual Twin Tower Translation Language Models (T3LM) for e-Commerce Product Search
Vijay Huddar
Sr. Applied Scientist, Amazon Search
Bangalore
Vijay Huddar is a Sr. Applied Scientist at Amazon Search, presently working on improving the worldwide search quality for the secondary languages. He has been part of Amazon for more than 5 years. As part of the ML team, Vijay has worked extensively on Semantic Search, Causal Inference, and Delivery experience. Prior to Amazon, Vijay had 2.5 years stint with Xerox Research Centre India, building ML algorithms for the healthcare domain. Vijay’s research has led to multiple patents and papers.
Leveraging Road Network Context to Solve Last Mile Stop Consolidation
Sai Krishna Tejaswi Nimmagadda
Applied Scientist, Amazon Last Mile
Bangalore
Sai Krishna Tejaswi Nimmagadda is an Applied Scientist at Amazon Last Mile Science team. Prior to Amazon he worked as a Strats Associate in Goldman Sachs. Prior to that he worked as an Academic Researcher in Machine Learning domain at UC Irvine. He has done Bachelors from IIT Kharagpur and post his graduation he has worked as a Software Engineer at Ericsson & Verizon. He has worked in multiple domains such as NLP, Computer Vision and at Amazon working in Geospatial domain.
A decision-tree framework to select optimal box-sizes for product shipments
Karthik Gurumoorthy
Sr. Applied Scientist, Amazon Search
Bangalore
Karthik Gurumoorthy graduated with a dual masters degree in Mathematics and CS and did his PHD in Computer Science from the University of Florida. He worked at GE Global Research for 3 years in the field of medical image analysis. He spent 1.5 years at the ICTS-TIFR, Bangalore where he conducted research in data assimilation and filtering theory. At present, he is a Senior Applied Scientist working on domains of causal inference, density estimation, filtering theory, and signal processing.
Solving Price Per Unit Problem Around the World: Formulating Fact Extraction as Question Answering
Kushal Kumar
Applied Scientist, Amazon ML
Bangalore
Kushal Kumar is Applied Scientist at IML. Graduate from IIT Kanpur with major in Mathematics and Scientific Computing. Prior to Amazon, he worked at Goldman Sachs and at IBM Research as a Research Intern in Image Processing domain. Kushal has a deep interest in different fields of machine learning and pattern recognition, and have worked on Semi-Supervised GANs, Fixing Degraded Facial Images using GANs, predicting short story endings attribute extraction from unstructured text, among others.
Discovery and predictions for molecules and crystalline materials using graph based DL models
Niloy Ganguly
Professor, IIT Kharagpur
Kharagpur
Dr. Niloy Ganguly is a Professor in the Dept. of Computer Science and Engineering at IIT Kharagpur and a Fellow of Indian Academy of Engineering. He is presently a visiting professor in Leibniz University of Hannover. His research interests lie primarily in Social Computing, Machine Learning, and Network Science. He has published in 60 journals and 160 conferences such as NeurIPS, KDD, ICDM, IJCAI, WWW, CSCW, EMNLP, CHI, ICWSM, INFOCOM, Physical Reviews, IEEE and ACM Transaction.
AI in Alexa
Shankar Ananthakrishnan
Director, Amazon Alexa
Cambridge, MA
Shankar Ananthakrishnan has two decades of experience in speech recognition, NLU, speect to speech translation etc. He received his Ph.D. in Electrical Engineering from the University of Southern California, and is currently a Director of Applied Science, Alexa AI at Amazon. Previously, he was a Senior Scientist at Raytheon BBN Technologies. He has published over 50 papers, and is the recipient of best paper awards at leading conferences, including ICASSP and Interspeech.
Early Warning and Early Intervention
Nina Mishra
Principal Applied Scientist, Amazon AWS
East Palo Alt
Nina Mishra is an experienced scientist, inventor and author with more than 50 publications and 15 awarded patents. She enjoys creating fast algorithms that discover ML-based insights on streaming, ever-changing, big data. Her aspiration is to be the mastermind behind more ML solutions in the healthcare domain: to envision new applications, to technically lead the effort, to build the solution, and to positively impact the lives of tens of millions of people.
Scaling Natural Language Processing for the Next Billion Users
Partha Pratim Talukdar
Staff Research Scientist
Indian Institute of Science & Google
Partha is a Staff Research Scientist at Google Research, Bangalore where he leads a group on Natural Language Understanding. He is also an Associate Professor (on leave) at IISc Bangalore. Partha is broadly interested in Natural Language Processing, Machine Learning, and Knowledge Graphs.
Ad Headline Generation using Self-Critical Masked Language Model
Yashal Kanungo
Applied Scientist, Amazon Ads
Bangalore
Yashal is an Applied Scientist with interests in automated generation of content, information retrieval and a variety of other NLP tasks. He has also worked extensively on deployment and working of ML models in production and have additional experience with large scale (10s of TBs at a time so far) feature engineering and analytics.
Multilingual Product Search using Product Images for Semantic Alignment
Sourab Mangrulkar
Applied Scientist, Amazon India ML
Bangalore
Sourab Mangrulkar is part of the India Machine Learning team at Amazon. He works in the domain of Sponsored Products pertaining to the problems related to Sourcing, Relevance and Ranking. His research interests lie in Natural Language Processing and Computer Vision.
Scalable Text Understanding from Product Images
Yang Liu
Applied Scientist, Amazon India ML
Seattle
Yang Liu obtained his PhD in Information Science from University of Michigan in 2016. He joined Amazon as an applied scientist in 2018 after working in a startup for one and half year. He has been working on multiple projects in pricing and product catalog quality in Amazon. He is broadly interested in data mining, NLP, machine learning and their applications in E-commerce and health domain.
Squeezing the last DRiP: AutoML for cost-constrained Product classification
Abhisek Divekar
Research Engineer, Amazon India ML
Bangalore
Abhishek Divekar is a Research Engineer - II at the India Machine Learning department at Amazon. His work primarily focuses on AutoML and NLP, with an emphasis on automatic methods for cost-optimization and text augmentation. He is completing his Masters from UTexas, Austin and completed his bachelors from VJTI, Mumbai.
Deep Declarative Networks with Application to Blind PnP
Stephen Gould
Amazon Scholar, Amazon India ML
Bangalore
Stephen Gould is a Professor of Computer Science at the Australian National University (ANU), Australian Research Council (ARC) Future Fellow and Amazon Scholar. He received his PhD in Computer Science and Electrical Engineering from Stanford University in 2010. Stephen has broad interests in the areas of computer and robotic vision, machine learning, deep learning, structured prediction, and optimization.
Closing Remarks
Anton van den Hengel
Director, Amazon India ML
Adelaide
Anton van den Hengel is a Director of Applied Science at Amazon, the Director of the Centre for Augmented Reasoning at the Australian Institute for Machine Learning (AIML), a Professor of Computer Science at the University of Adelaide, a Fellow of the Australian Academy of Technology and Engineering. Anton was the founder of AIML, which is Australia’s largest machine learning research group, and currently number 2 Computer Vision research group globally by publications (see csrankings.org).
Posters
A Data Bootstrapping Recipe for Low-Resource Multilingual Relation Classification
Arijit Nag | IIT Kharagpur
Arijit Nag is a doctoral student at Computer science & Engg dept., IIT Kharagpur, working with Prof. Animesh Mukherjee and Prof. Niloy Ganguly. He is interested in working on real-life problems in natural language processing, focusing mainly on problems with limited resources.
Causal contextual bandits with targeted interventions: a new method for smarter experimentation
Chandrasekar Subramanian | IIT Madras
Chandrasekar Subramanian is currently a PhD student at IIT Madras working with Prof. Balaraman Ravindran, and his research focuses on causality and reinforcement learning. He is a member of the Robert Bosch Center for Data Science & AI. He was previously working with Microsoft, before which he was in consulting roles at McKinsey and Oliver Wyman, and in a research scientist role at Tata. He holds an MTech (Computer Science) from IIT Madras and an MS (Financial Economics) from the University of Oxford.
NAGAR: Resolving Addresses to Neighbourhoods for Efficient Delivery Planning in Emerging Marketplaces
Govind | Last Mile ML Science
Govind is an Applied Scientist II at the Last Mile ML Science team in Hyderabad. He has been working on the customer address related projects, more recently project NAGAR which involves learning neighbourhoods from customer addresses and further resolution of addresses to appropriate neighbourhoods. He received his PhD from University of Caen, France and later worked as a post-doctoral researcher before joining Amazon. He has worked on multiple research problems in the domains of NLP and ML, such as event spread prediction among communities on the Web, building noise-resilient contextual word embeddings, and so on.
Self-supervised Representation Learning for User Ad Activity Sequences
Rajat Agarwal | Amazon
Rajat is an Applied Scientist in the Traffic Quality team at Amazon Ads where he works on building large-scale and low-latency deep learning techniques for robot and ad fraud detection. His research interests include semi and self-supervised learning, generative modeling, model-free reinforcement learning and their applications in computer vision, natural language processing, time-series modeling and anomaly detection. He has graduated from BITS Pilani Goa Campus in 2018 with a major in computer science
Improving Geocode Learning for AMZL through Address Normalisation
Saket Maheshwary | Amazon
Saket Maheshwary joined Amazon as a scientist with the Last Mile Science team in April 2019. He has five years of overall industry experience. He explored his research interests during his master's at IIIT Hyderabad and works at the intersection of data mining, machine learning, deep learning, and natural language processing. He has published his research at some of the top artificial intelligence conferences in the world. Besides being responsible for researching some of the challenging problems related to these areas, he has also productized the research insights to real-world solutions.
Responsible NLU Cost-effective bias identification and data-sheets for NLU models
Satyam Dwivedi | Amazon
Satyam Dwivedi is a Language Engineer in Alexa-AI NLU team which bootstraps and maintains NLU models for Indian English and Hindi languages. I hail from Kannauj, completed my undergrad and postgrad degrees from BHU, and currently pursuing my PhD in Computational Linguistics from IIT BHU. My research interests are formal semantics, syntactic models, conversational and responsible AI.
Thinking Beyond Complete Data with Deep Learning and Time Series
Vinayak Gupta | IIT, Delhi
Vinayak is a Ph.D. student in the department of computer science, I.I.T. Delhi, under the supervision of Prof. Srikanta Bedathur. His research interests are in designing neural graph and temporal point processes for sequential data mining. Previously, he received his B.Tech in computer science from I.I.I.T. Jabalpur
Building Data-Efficient High-Performance Product Classifiers
Vinayak Puranik | Amazon
Vinayak is a Sr. Applied Scientist with India Machine Learning team at Amazon. He has worked on multiple impactful ML initiatives at Amazon such as, high-performance product classification for multiple business use-cases, AutoML solutions and improving catalog quality. Vinayak has a Master’s degree in Computer Science from IISc, Bangalore.
Learning Neural Models for Combinatorial Problems
Yatin Nandwani | IIT, Delhi
Yatin Nandwani is a PhD scholar in the area of Machine Learning and Artificial Intelligence, guided by Mausam and Parag Singla at Computer Science and Engineering Department, Indian Institute of Technology Delhi. Prior to joining the PhD program, worked in the quantitative finance industry for 10 years. Did my graduation in Mathematics and Computing, a five year integrated M.Tech programme offered by Mathematics Department at IIT Delhi.
Robust non‐parametric regression via incoherent subspace projections
Bhaskar Mukhoty | IIT Kanpur
Bhaskar Mukhoty is a PhD student in the department of CSE, IIT Kanpur, advised by Dr. Purushottam Kar and Prof. Sandeep Kumar Shukla. The focus area of his PhD thesis is robust learning, i.e. learning estimator/function with reasonable accuracy from corrupted training data. They are inclined towards easy to implement algorithms that work under adversarial corruption with provable guarantee and also interested in applications of machine learning in security and time-series forecasting.
Learning Language Agnostic Models Exploiting Multi-modal Knowledge Distillation
Akanksha Paul & Gaurav Rajput | Amazon
Akanksha Paul loves reading, coding and solving new problems and have an avid interest in Machine Learning and Computer Vision. Currently working as a Applied Scientist in MARS team.
Gaurav works in Amazon as an Applied Scientist in Advertising domain. He enjoys solving complex problems with the help of Deep learning both in Computer Vision and NLP.
Multilingual Open Information Extraction
Keshav Kolluru | IIT Delhi
Keshav is a PhD student at IIT Delhi working on the topic of Open Information Extraction, under the guidance of Prof. Mausam and Prof. Soumen Chakrabarti. He has completed successful internships at Google Research, IBM Research and Microsoft and has a BTech degree from IIT Bhubaneswar.
Can Non-Humanoid Social Robots Reduce Workload of Special Educators (Accepted at ICRA 2021)
Nabanita Paul | CSA department, IISc
Nabanita Paul is a PhD student in Machine Learning lab in CSA department, IISc. Her current interest is designing algorithms that allow Social Robots to offer personalized assistance to Special Educators in autism education, and in general, other neuro-developmental disorders.
FAX: Free-text Attribute eXtraction
Nitesh Methani | Amazon
Nitesh Methani started my career at Amazon in Jun 2020 when he joined Last Mile Science as an Applied Scientist where he has worked on building a ML model that extracts business contextual attributes from free-text delivery instructions provided by our customers. Apart from scaling this model to multiple geographies, He has been exploring multiple address related projects including address text normalization and address parsing. He holds a Master’s degree from IIT Madras with specialization in Deep Learning and have published my research in AAAI, WACV, and AMLC. Outside of work he also do a small number of things on repeat: cycling, reading biographies, DIY investing books, and watching science fiction movies.
NeuroMLR: Robust & Reliable Route Recommendation on Road Networks
Sahil Manchanda | IIT, Delhi
Sahil Manchanda is a PhD Scholar at the Department of Computer Science and Engineering at the Indian Institute of Technology Delhi. He is working under the supervision of Prof.Sayan Ranu. Sahil’s research is spanned across the broad area of Learning algorithms over graphs with a particular focus on 'Learning heuristics for Combinatorial Optimization problems' and 'Large scale machine learning on graphs'. Prior to joining the PhD program, he has worked as a Research Engineer at Conduent Labs, erstwhile Xerox Research Center India. He has obtained his master’s degree from the Indian Institute of Technology Guwahati
Clustering-based Distributed KNN System
Yikai Ni | Amazon
Yikai Ni is a software engineer at Amazon. He is experienced in building large-scale distributed systems, applications of machine learning to similarity recommendation and other related problems. In Amazon, he is instrumental in designing and implementing a general product similarity service which has been widely used in the company, and led the improvement of the similarity search component to support over 500MM input by adding the distribution logic. Prior to joining Amazon, he received his master's degree in Computer Science from University of Southern California.
Robust Machine Learning
Pankaj Kumar Sharma | Amazon
Pankaj Kumar Sharma is working as Research Scientist II in Amazon Alexa NLU team which takes care of Natural language understanding models for two i18 locales : Indian English and Hindi. I hail from Jaipur, graduated from IIT Delhi and then built, hustled, enjoyed and failed few start-ups: Langhar.com and Yumist. Previously, I have also worked on automated speech recognition, NLP and computer vision to drive business growth in Policy bazaar.
Confused about Efficient Transformer Models? This Survey is All you Need
Ankith M S | Amazon
Ankith M S is a Machine Learning Scientist at IML, Amazon. I have been with Amazon for four years, before which I helmed a startup named Mapicle, incubated at SINE, IIT-B, where we worked on building hierarchical summarization of news. I completed MTech from CSE, IIT Bombay. My specialization was and is NLP. At Amazon, I have mostly worked on Sponsored Ad projects focussing on query reformulation, semantic sourcing, relevance models and eCTR.