Amazon Research Awards

Call for proposals

The 2017 Amazon Research Awards application period closed on September 15, 2017, and the winners were announced on January 15, 2018. The 2018 Amazon Research Awards call for proposals and application period will be announced in June 2018.

Submission Requirements

Project proposals should be a maximum of 4 pages (single column, minimum 10 pt font), plus 1 page for references, plus CV. All content components (proposal, references, CV) should be composed into a single PDF file.

Project proposals must include:

  1. Motivation for a specific research task.
  2. An outline of the approach, challenges and expected results.
  3. A proposed budget not exceeding $80,000 in cash and $20,000 in AWS Promotional Credits.
  4. CV of the leading faculty member. CV can be submitted as a link listed at the end of the proposal document.

Computer vision

  • Apparel similarity
  • Hand and head gesture recognition, pose estimation
  • Image captioning
  • Large scale content based image retrieval
  • Low-level vision and image processing
  • Motion: segmentation, tracking (moving camera), gait recognition
  • Recognition: detection, categorization, segmentation and matching
  • Robot vision
  • Structure-from-motion, stereo and 3D reconstruction
  • Video: events and activities

General AI

  • Autonomous systems
  • Knowledge representation
  • Planning
  • Reasoning and decision making

Knowledge management and data quality

  • Data cleaning for machine learning
  • Knowledge extraction from the web
  • Knowledge verification from external sources
  • Large-scale data alignment and cleaning
  • Leveraging structured knowledge in deep learning
  • Personalization using personal knowledge base
  • Product trend discovery from web
  • Quantitative and logical error detection

Machine learning

  • Active learning
  • Continuous Learning
  • Data and resource efficient learning
  • Deep learning
  • Parallel and Distributed Learning
  • Privacy preserving learning
  • Reinforcement learning
  • Representation learning
  • Transfer learning

Machine translation

  • Advances in methods for estimating MT quality at run time
  • Advances in MT for noisy and user-generated content
  • Advances in scalable real-time neural MT
  • Efficient training for neural MT
  • Incremental adaptation of neural MT systems
  • Learning from weak user feedback
  • Multitask learning for MT
  • Reinforcement learning for MT
  • Using document context and other modalities to improve MT

Natural language understanding

  • Advanced synonym and hypernym generation for eCommerce search
  • Deep semantic representations for eCommerce queries (both keyword and natural language queries)
  • Fact extraction from unstructured data
  • Fact verification and trustworthiness of text sources
  • Named entity recognition, linking and resolution
  • Null and sparse result set recovery for eCommerce search (includes query relaxation)
  • Question answering


  • Approaches to recommend products that uses customer's fashion preferences
  • Approaches to recommend products without behavior data ("cold start" item recommendations)
  • Content based approaches for estimating popularity of newly or yet-to-be released products
  • Item-to-item collaborative filtering using deep learning
  • Learning user preferences over ontologies/interests
  • Online experimentation approaches for selecting best recommendation strategies


  • Adaptive, online learning for item identification
  • Cloud-based planning and control for mobile robots
  • Detection, tracking, and trajectory prediction of dynamic objects
  • Explaining or debugging decisions made by complex algorithms
  • Grasping for reliability and robustness
  • Human-in-the-loop teleoperation
  • Legible intent and social movement of robots around people
  • Many-robot coordinated motion planning
  • Map building based on data collected by a robot
  • Persistent and robust 3D SLAM in changing environments
  • Segmentation and identification in dense visual scenes or with minimal specific training
  • Simulation of sensing and grasping for object manipulation
  • Stochastic modeling, control, and optimization of automated warehouses
  • Task allocation or partitioning for mobile robots

Search and information retrieval

  • Advances in ranking for tail (infrequent) eCommerce queries
  • Document quality techniques for search ranking features and spam detection
  • Index tiering based on advances in document quality assessment
  • Product Knowledge Graph integration into eCommerce search algorithms
  • Semi-automatic Product Knowledge Graph and navigation taxonomy construction

Security, privacy and abuse prevention

  • Bot detection
  • Browser/device fingerprint and digital forensics  
  • Early detection of emerging patterns with limited labeled data (one-shot-learning)
  • Graph modeling (latent representations from a graph and anomaly detection)
  • Human-in-the-loop machine learning
  • Web behavioral modeling, online identity and password-less authentication


  • Advanced deep learning methods for speech recognition
  • ASR error detection and confidence estimation
  • Cross-lingual transfer learning
  • Emotion recognition
  • Keyword spotting
  • Media and recorded speech detection
  • Multi-lingual ASR
  • Real-time ASR in embedded/low resource applications
  • Robustness to microphone variation, microphone independent modeling
  • Speaker diarization
  • Speech recognition and speech end-pointing robustness to background speech, music, noise, reverberation 
  • Spoken dialog systems and conversational understanding
  • Unsupervised training, adaptation in automatic speech recognition (ASR)
  • Very large vocabulary, domain-independent ASR


Full-time faculty members of institutions in North America and Europe granting PhD degrees in fields related to Machine Learning are eligible to apply.

Awards Structure

Awards are structured as one-year unrestricted gifts to academic institutions. Though the funding is not extendable, applicants can submit new proposals for subsequent calls.

Award Process

Project proposals are reviewed by an internal awards panel and the results are communicated to the applicants approximately three months after the submission deadline. Each project will be assigned an Amazon researcher contact. The university researchers are encouraged to maintain regular communication with this contact to discuss ongoing research and project progress. Researchers are also encouraged to publish the outcome of the project and commit any related code to open-source code repositories.