Skip to main content

Main navigation

  • About BCS
    • Mission
    • History
    • Building 46
      • Building 46 Room Reservations
    • Leadership
    • Employment
    • Contact
      • BCS Spot Awards
      • Building 46 Email and Slack
    • Directory
  • Faculty + Research
    • Faculty
    • Areas of Research
    • Postdoctoral Research
      • Postdoctoral Association and Committees
    • Core Facilities
    • InBrain
      • InBRAIN Collaboration Data Sharing Policy
  • Academics
    • Course 9: Brain and Cognitive Sciences
    • Course 6-9: Computation and Cognition
      • Course 6-9 MEng
    • Brain and Cognitive Sciences PhD
      • How to Apply
      • Program Details
      • Classes
      • Research
      • Student Life
      • For Current Students
    • Molecular and Cellular Neuroscience Program
      • How to Apply to MCN
      • MCN Faculty and Research Areas
      • MCN Curriculum
      • Model Systems
      • MCN Events
      • MCN FAQ
      • MCN Contacts
    • Computationally-Enabled Integrative Neuroscience Program
    • Research Scholars Program
    • Course Offerings
  • News + Events
    • News
    • Events
    • Recordings
    • Newsletter
  • Community + Culture
    • Community + Culture
    • Community Stories
    • Outreach
      • MIT Summer Research Program (MSRP)
      • Post-Baccalaureate Research Scholars
      • Conferences, Outreach and Networking Opportunities
    • Get Involved (MIT login required)
    • Resources (MIT login Required)
  • Give to BCS
    • Join the Champions of the Brain Fellows Society
    • Meet Our Donors

Utility Menu

  • Directory
  • Apply to BCS
  • Contact Us

Footer

  • Contact Us
  • Employment
  • Be a Test Subject
  • Login

Footer 2

  • McGovern
  • Picower

Utility Menu

  • Directory
  • Apply to BCS
  • Contact Us
Brain and Cognitive Sciences
Menu
MIT

Main navigation

  • About BCS
    • Mission
    • History
    • Building 46
    • Leadership
    • Employment
    • Contact
    • Directory
  • Faculty + Research
    • Faculty
    • Areas of Research
    • Postdoctoral Research
    • Core Facilities
    • InBrain
  • Academics
    • Course 9: Brain and Cognitive Sciences
    • Course 6-9: Computation and Cognition
    • Brain and Cognitive Sciences PhD
    • Molecular and Cellular Neuroscience Program
    • Computationally-Enabled Integrative Neuroscience Program
    • Research Scholars Program
    • Course Offerings
  • News + Events
    • News
    • Events
    • Recordings
    • Newsletter
  • Community + Culture
    • Community + Culture
    • Community Stories
    • Outreach
    • Get Involved (MIT login required)
    • Resources (MIT login Required)
  • Give to BCS
    • Join the Champions of the Brain Fellows Society
    • Meet Our Donors

Events

News Menu

  • News
  • Events
  • Newsletters

Breadcrumb

  1. Home
  2. Events
  3. Entity Abstraction in Visual Model-Based Reinforcement Learning
Department of Brain and Cognitive Sciences (BCS)
Seminar

Entity Abstraction in Visual Model-Based Reinforcement Learning

Speaker(s)
Michael Chang
Add to CalendarAmerica/New_YorkEntity Abstraction in Visual Model-Based Reinforcement Learning02/28/2020 8:00 pm02/28/2020 10:00 pm46-3002 | Singleton Auditorium
February 28, 2020
8:00 pm - 10:00 pm
Location
46-3002 | Singleton Auditorium
Contact
Sholei Croom
    Description

    Many robotic tasks are specified in terms of objects and their relationships, so naturally a learner that also represents a physical scene in terms of distinct objects and relations would enjoy broader generalization and more efficient planning than a learner that computes a monolithic global encoding of the sensory observations. However, for robots that learn from raw video data, the difficulty of grounding object representations in raw pixels without domain-specific supervision has led much work in end-to-end robotic learning to use monolithic global encodings, which require less domain-specific supervision but produce entangled representations that makes generalization and planning less straightforward. In particular, a seemingly trivial but important property we desire in our models is that they can transfer what they learn about a single object across different contexts: the functions that govern the appearance and dynamics of objects should not be affected by the number or configuration of other objects in the scene, yet this type of generalization is not straightforward to enforce with monolithic global encodings because these encodings generally entangle information about all objects together.

     

    In this talk, I will present a model that combines the advantages of representing objects and relations with the advantages of learning from raw data: it processes a factorized latent representation that enforces the variable isolation required for transferring knowledge about a single object across different contexts, and it also learns entirely end-to-end from raw pixels without any manual specification about what objects are and what the factors in the representation should correspond to. 

    It learns to ground the latent factors on actual objects in the visual observation by executing an inference procedure on a probabilistic entity-factorized dynamic latent variable model.

    I will show on both simulated block stacking problems and real robot interaction video data that many desirable properties for physical modeling and planning — such as temporal continuity, handling occlusion, unsupervised segmentation, and symbolic grounding  — emerge  in some form as direct consequences of the architecture of the model and the algorithm we use to train it, suggesting a promising direction for enabling better generalization and planning in real learning robotics.

    For more information please see: https://arxiv.org/abs/1910.12827.

    Speaker Bio

    Michael Chang is a Ph.D. student at U.C. Berkeley advised by Professors Sergey Levine and Tom Griffiths. He is broadly interested in designing machine learners that generalize in similar ways that humans do. In his research he seeks to understand how to enable neural learners to automatically instantiate, compose, and abstract their own primitive computations and representations in a way that can be used to accelerate future learning and improvise solutions to novel problems. He received his B.S. in EECS from MIT, where his undergraduate research was supervised by Professor Joshua Tenenbaum and Antonio Torralba. Before entering Ph.D., he has interned with Professors Honglak Lee and Jürgen Schmidhuber. He is a recipient of the NSF Graduate Fellowship. For additional information, please see: http://mbchang.github.io/.

    Upcoming Events

    See All Events
    Don't miss our next newsletter!
    Sign Up

    Footer menu

    • Contact Us
    • Employment
    • Be a Test Subject
    • Login

    Footer 2

    • McGovern
    • Picower
    Brain and Cognitive Sciences

    MIT Department of Brain and Cognitive Sciences

    Massachusetts Institute of Technology

    77 Massachusetts Avenue, Room 46-2005

    Cambridge, MA 02139-4307 | (617) 253-5748

    For Emergencies | Accessibility

    Massachusetts Institute of Technology