Weekly Research Meetings

CBMM Weekly Research Meeting

Jan 13, 2016 - 4:00 pm
Venue:  MIT Bldg. 46 Room TBA Speaker/s:  Tomaso Poggio, Qianli Liao, Georgios Evangelopoulos, and Tejas Kulkarni.

Thrust 5: Theories of Intelligence 

The thrust aims to provide theoretical frameworks and common mathematical tools for understanding visual intelligence, for guiding computer implementations, and for informing and interpreting experiments in the Center.

Our working hypothesis for vision suggests two main different stages of visual processing:

  • The first 100ms of vision in the ventral stream are mostly feedforward and correspond to what we may call immediate perception (following Julesz). We conjecture that the main computational goal is to generate image representations that can be used to answer different types of basic questions. Our theoretical framework  to describe feedforward processing in the hierarchy of cortical areas is i-theory.
  • More specific, task-dependent questions require different computations that involve top-down, possibly iterative processing. We are working on two theoretical frameworks  to deal with this stage: generative models&probabilistic inference and  top-down visual routines.

We will give a brief overview of the state of Thrust 5, followed by flash-like presentations of three projects:

  1. application of i-theory to the face network in macaque brain
  2. application of i-theory to speech
  3. examples of models based on the probabilistic/generative framework

We will then open an informal discussion around the approach, its problems and its many open questions.

CBMM Weekly Research Meeting

Dec 16, 2015 - 4:30 pm
Tangetal
Venue:  MIT Bldg. 46 Room 5165 Address:  5th Floor, MIT Bldg 46, 43 Vassar St., Cambridge MA 02139 Speaker/s:  Gabriel Kreiman

Thrust 2 Projects

Hanlin Tang and Bill Lotter - Pattern completion: behavior, physiology and computation

Matt Wilson - Presentation of the Penagos-Gershman research collaboration

Ed Boyden: Technologies for Mapping the Circuits and Mechanisms of Intelligence

Please note change in start time, this meeting will start at 4:30pm.

CBMM Weekly Research Meeting

Dec 2, 2015 - 4:30 pm
CBMM logo
Venue:  Harvard Northwest Building, Room 243 Address:  52 Oxford St., Cambridge, MA 02138 Speaker/s:  Shimon Ullman, Boris Katz

Thrust 3 Projects

Yevgeni Berzak - Human Language Learning
Abstract: Linguists and psychologists have long been studying cross-linguistic transfer, the influence of native language properties on linguistic performance in a foreign language. In this work we provide empirical evidence for this process in the form of a strong correlation between language similarities derived from structural features in English as Second Language (ESL) texts and equivalent similarities obtained from the typological features of the native languages. We leverage this finding to recover native language typological similarity structure directly from ESL text, and perform prediction of typological features in an unsupervised fashion with respect to the target languages. Our method achieves results that are highly competitive with those obtained using equivalent methods that rely on typological resources. Finally, we outline ongoing work that utilizes transfer phenomena for prediction of grammatical error distributions and syntactic parsing for learner language.

Andre Barbu - Disambiguation and grounded Question Answering
Abstract: Humans perform many language tasks that are grounded in perception: description, language acquisition, disambiguation, question answering, etc. These tasks share common underlying principles and we show how a single mechanism can be repurposed to address many of them. In disambiguation we show have an ambiguous sentence with multiple interpretations can be understood in the context of a video depicting one of those interpretations. In question answering we show have a question can be answered given a video by repurposing a vision-language sentence generation mechanism. Our methods share data and do not need to be retrained for every novel task. We briefly discuss extensions to other vision-language tasks and to language-only tasks such as translation that can be potentially solved through perception.

Candace Ross - Predicting the Future
Abstract: We present a model which predicts the motion of objects in still images. This predictive ability is important to humans; our reflexes are relatively slow and if we were limited to always responding to the information at hand we would be unable to navigate complex environments and avoid dangers. Our goal is to bring to bear an insight from cognitive science to vision: humans can reliably predict to motion of objects from still images. We demonstrate how a model can learn this ability in an unsupervised fashion by observing YouTube videos. Our model employs a deep neural network to predict dense optical flow given single frames. This approach is competitive with standard optical flow algorithms such as Lucas-Kanade, which require videos not just images. This work provides a new foundation for many techniques such as learning to automatically segment objects, improved robotic navigation, extending video action recognition to images, and the development of a cognitively-motivated attention mechanism.

Presentation: CBMM Thrust 3 research projects

CBMM Weekly Research Meeting: Imaginative Reinforcement Learning

Nov 18, 2015 - 4:00 pm
CBMM logo
Venue:  MIT McGovern Institute for Brain Research (MIT Bldg 46) Address:  43 Vassar St., Cambridge MA 02139 MIT Bldg 46, 5th Floor, MIBR Reading Room #46-5165 Speaker/s:  Sam Gershman  

Abstract: Reinforcement learning is typically conceived of in terms of how reward predictions and choice behavior adapt based on an agent's experience. However, experience is too limited to provide the brain with the knowledge necessary for adaptive behavior in the real world. To go beyond experience, the brain must harness its imaginative powers. Applications of imagination to reinforcement learning include prospective simulation for planning, and learning cached values from imaginative episodes. I will discuss how these ideas can be formalized, recent experimental evidence, and connections to other ideas being explored in CBMM.

Gershman Lab website: http://gershmanlab.webfactional.com/index.html

CBMM Weekly Research Meeting: Thrust 1 Projects

Nov 4, 2015 - 4:00 pm
CBMM logo
Venue:  Harvard Northwest Building, Room 243 Address:  52 Oxford Street, Cambridge MA Speaker/s:  Josh Tenenbaum

Tomer Ullman - 1) Whether and how infants understand unified notions of mass and force (central to the question of how babies understand intuitive physics) and 2) Do children understand 'cost' in a utility calculation as related to physical effort.  

Julian Jara-Ettinger - "Costs, rewards, and commonsense psychology" Abstract: Humans have a unique ability to reason about other people’s unobservable mental life. By simply watching someone’s actions we can infer what they think and what they want, we can predict how they might act in the future, and we can judge their actions as praiseworthy or condemnable. In this three minute presentation I will propose the hypothesis that at the core of social cognition is a naïve utility calculus that enables us to reason about how agents behave as a function of the costs and rewards associated with different outcomes.

Yibiao Zhao - "Intuitive physics and psychology for perceiving, reasoning and learning about visual scenes"

CBMM Weekly Research Meeting

Oct 7, 2015 - 4:15 pm
Venue:  Harvard Northwest Building Room 255 Address:  52 Oxford St, Cambridge MA Speaker/s:  Tomaso Poggio

Agenda: 

(1) NIPS 2015 CBMM events

(Max Nickel): Symposium: Brains, Minds and Machines
(Tejas Kulkarni): Workshop on Black Box Learning and Inference
(Marco Cusumano-Towner): Workshop on Bounded Optimality and Rational Metareasoning

(2)  Future CBMM workshops (Turing++) (Tomaso Poggio)

(3) CBMM open publication model (Tomaso Poggio)

 

CBMM Weekly Meeting: Full interpretation of minimal images & the state of CBMM

Sep 30, 2015 - 4:15 pm
minimal image pairs
Venue:  MIT MIBR 46-3189 Speaker/s:  Guy Ben-Yosef:  Full interpretation of minimal images

Abstract:

We aim to model the process of ‘full interpretation’ of object images, which is the ability to identify and localize all semantic features and parts that are recognized by human observers. We model the interpretation process by identifying primitive components and relations that play a useful role in full interpretation by humans. To identify useful components and relations used in the interpretation process, we consider  the interpretation of 'minimal configurations’, namely reduced local regions that are minimal in the sense that further reduction will turn them unrecognizable and uninterpretable. We describe experimental results of our model, and discuss approaches of interpretation modeling to recognize human activities and human interactions, which are beyond the scope of current models of visual recognition.

CBMM Weekly Research Meeting: The Future of AI: Opportunities and Challenges

Mar 3, 2015 - 4:00 pm
Max Tegmark
Speaker/s:  Max Tegmark

The Future of AI: Opportunities and Challenges

Abstract: I will give a report from the recent Puerto Rico conference on this topic (http://futureoflife.org/misc/ai_conference) and lead a discussion of what we can do today to help maximize the future benefits of AI while avoiding pitfalls.

 

CBMM Weekly Research Meeting: Fast reading of actions in a competitive interaction

May 12, 2015 - 4:00 pm
Fast reading actions image
Venue:  MIT 46-3189 Address:  43 Vassar St, Cambridge, 02139 Speaker/s:  Speaker: Maryam Vaziri Pashkam  

Abstract

Humans are experts at reading others’ actions. They efficiently process others’ movements in real time to predict intended future movements. Here we designed a competitive reaching task to investigate real-time body reading in a naturalistic setting. Two subjects faced each other separated by a plexiglass screen. Fingertip positions were recorded with magnetic sensors. One subject (Attacker) was instructed to tap one of two targets on the screen and the other subject (Blocker) was told to reach for the same target as quickly as possible. Reaction times, measured as the difference in the initial finger movement of the Attacker and the Blocker were fast, much faster than reaction times to a moving dot projected on the screen. This suggests Blockers use preparatory actions of Attackers to predict their goal. To further confirm the role of preparatory cues we video-taped an Attacker and systematically removed the preparatory information from the videos and showed that reaction times are slower in response to the manipulated videos. To localize the position of preparatory information in the body we occluded various body parts of the Attacker. Reaction times got slower only when most of the body of the Attacker was occluded suggesting that preparatory cues are distributed over the body of the Attacker. Finally we presented the Blockers with only the preparatory information and asked them to explicitly report the Attacker’s intended target. Accuracies in this explicit report were lower compared to the accuracies in the arm movement task suggesting implicit access to preparatory cues during arm movements. Taken together, these results suggest participants efficiently and implicitly gather preparatory cues from the body of their partner to predict their intentions well before movement begins.

Pages

Subscribe to Weekly Research Meetings