Weekly Research Meetings

CBMM Weekly Research Meeting: Computing scale and translation invariant representations requires an eccentricity-dependent cortical magnification factor.

Apr 4, 2014 - 4:00 pm
Venue:  Harvard University: Northwest Bldg, Room 243 Address:  52 Oxford Street, Harvard University Northwest Building, Cambridge, 02138 Speaker/s:  Tomaso Poggio and Jim Mutch; CBMM Thrust 5 – Theories for Intelligence

Progress of CBMM Challenge – Enabling Theory

Abstract:

We continue the series of weekly discussions and reports on each CBMM challenge question [e.g. What is there? What will happen next? What are they doing? etc.] describing progress and problems of ongoing work at CBMM.

This Friday we will speak informally about very preliminary work with the explicit goal of brainstorming about where to go. Basic properties of recognition at a glance are predicted by a sampling for magic theory. The emerging picture is consistent with Ullman’s ideas on minimal images and implies that recognition under natural conditions happens by composing information from a set of fixations, with each fixation providing recognition from image patches of about arrays of 30 by 30 pixels at different resolutions.
According to the magic theory, invariance to scale and translation predicts an architecture of visual cortex that is consistent with the data about cortical magnification factor. Our sampling extension of the theory suggests a size of minimal images of around 30 by 30, a fovea of size around 20’ at the highest resolution and translation invariance that depend linearly on spatial frequency.

Presenters hope there will be discussions on topics such as:
• Which psychophysical experiment should be done to clarify the current confusing literature on translation invariance in recognition?
• Can we predict crowding and Bouma’s law?
• Can we predict the cortical location of different recognition tasks?

CBMM Weekly Research Meeting: Progress of CBMM Challenge – Visual Understanding

Mar 21, 2014 - 4:00 pm
Visual Understanding: A view from the future
Venue:  Harvard University: Northwest Bldg, Room 243 Address:  52 Oxford Street, Harvard University Northwest Building, Cambridge, 02138 Speaker/s:  Andrei Barbu and Daniel Harari; CBMM Thrust 3 – Visual Intelligence

Progress on the CBMM challenge questions: What is there? and Who is there?

Title: Thrust 3: Vision and language

Abstract: A fundamental human ability is to communicate with others about what we are perceiving and to change our understanding of the world when others are communicating with us. To understand this capability, we (Thrust 3) take the CBMM Turing test literally by developing approaches to answering queries about images and videos.

Answering questions requires that we transfer knowledge between modalities. We start with a question posed in natural language, we connect our understanding of the question with the scene we are perceiving, we discover the answer to the question, and then take that perceptual knowledge and encode it back into language. The uncanny human ability to easily transfer knowledge between modalities points to the conjecture that human representations may be modality independent, and searching for such representations will produce approaches that are more cognitively plausible.

This week’s talk will focus on vision and how it connects to language. We will describe how representations in both the language and vision communities fall short of bridging the two. A new representation which helps bridge part of the gap between vision and language will be proposed. We will present our research ideas on question answering, language learning, and connections to other vision and language tasks which are key to human cognition.

CBMM Weekly Research Meeting: Hippocampus as a target structure for physiological examples of neural circuits solving problems.

Mar 14, 2014 - 4:00 pm
Venue:  MIT: McGovern Institute Reading Room, 46-5165 Address:  43 Vassar Street, MIT Bldg 46 , Cambridge, MA 02139 United States Speaker/s:  Greg Hale, Hector Penagos and Zhe Chen; CBMM Thrust 2 – Circuits for Intelligence

Progress on the CBMM challenge questions: Where did I come from? Where am I going?

Abstract: In the Jeopardy/Watson effort every week on Friday there was an evaluation of performance. We continue the series of weekly discussions and reports on each CBMM challenge question [e.g. what is there? what will happen next? what are they doing? etc] describing progress and problems of ongoing work at CBMM.

This week we propose to add two new  questions to the CBMM Challenge set: Where did I come from? Where am I going? We describe how the hippocampus my answer these questions. In particular, we will elaborate on the known spatial representations of the rodent hippocampus and explore how these representations can be used for solving problems. We will also present new findings that show how hippocampus interacts with cortex during periods in which rats disengage from ongoing tasks. We will discuss the possibility that this interaction reveals how hippocampus participates in cognitive functions such as planning, remembering the past and imagining the future. Lastly, we will present ongoing computational work that employs hidden Markov models to represent spatial trajectories encoded by hippocampal ensemble spiking patterns.

CBMM Weekly Research Meeting: The origin and development of uniquely human geometric intelligence

Feb 28, 2014 - 4:00 pm
Molly Dillon - photo credit Rose Lincoln
Venue:  Harvard University: Northwest Bldg, Room 243 Address:  52 Oxford Street, Harvard University Northwest Building, Cambridge, 02138

Progress on the CBMM challenge questions: what is there? and who is there?

Title: The origin and development of uniquely human geometric intelligence

Presenters: Moira Dillon, Elizabeth Spelke
Collaborators: L Mahadevan, Josh Tenenbaum; CBMM Thrust 1 – Development of Intelligence

Abstract: In the Jeopardy/Watson effort every week on Friday there was an evaluation of performance. We continue the series of weekly discussions and reports on each CBMM challenge question [e.g. what is there? what will happen next? what are they doing? etc] describing progress and problems of ongoing work at CBMM.

This Friday we will hear about the origin and development of uniquely human geometric intelligence. We will focus on studies with children and infants to explore what kinds of geometric information support uniquely human geometric knowledge, such as the appreciation of symbolic maps and intuitions about how distance and angle relate in triangles. Such knowledge develops in all humans regardless of formal schooling, but doesn’t come about until relatively late in a child’s development.

Presenters will initiate discussions on the following topics:
— What core concepts may be present from birth that support later geometric understanding?
— What learning mechanisms control this development?
— What role does experience play the the acquisition of these concepts? What kind of experience is necessary?
— How might we better elucidate the links between core geometric concepts and those that are later to emerge?

CBMM Weekly Research Meeting

Feb 14, 2014 - 4:30 pm
Venue:  Harvard University: Northwest Bldg, Room 243 Address:  52 Oxford Street, Harvard University Northwest Building Cambridge, MA 02138

Topic: Progress on the CBMM challenge questions: what is there?  and who is there?
Presenters: Camille Gomez and/or Gabriel Kreiman (work with Rick Born and JoJo Nassi).

Abstract:
In the Jeopardy/Watson effort  every week on Friday there was an evaluation of performance. We begin here a series of weekly discussions and reports on  each CBMM challenge question [e.g. what is there? what will happen next? what are they doing? etc] describing progress and problems of ongoing work at CBMM.
Today the discussion is about types of models  used for answering the what and who questions and in particular about whether they can be approximated as bottom-up architectures. Camille Gomez and Gabriel Kreiman will introduce the discussion by describing preliminary data from  V1 recordings after inactivating V2.

CBMM Weekly Research Meeting

Feb 7, 2014 - 4:30 pm
Venue:  MIT: McGovern Institute Reading Room, 46-5165 Address:  43 Vassar Street MIT Bldg 46 Cambridge, MA 02139 United States

Presentation: Research Thrust 2 – Circuits for Intelligence: Datasets and Metrics for the CBMM Challenge
Speakers: Prof. Gabriel Kreiman, Leyla Isik and Andrea Tacchetti

Abstract:
The performance of humans, computational models, or a population of neurons on a given task can be measured by single numbers such as ROC, but importantly, also by the consistency between their responses. For example, one can compare the correlation between the readout from a group of neurons and from a layer or module in a computational model.
In this meeting we will begin the important discussion of:
1) The various parts of an open database of videos and images to be used for the CBMM challenge over the next 5-10 years, and
2) Metrics that we will use to measure absolute performance of models, and their consistency with human behavioral performance and physiology.
We will discuss an example of preliminary work to compare model and human MEG performance on an action recognition task. In addition we will brainstorm datasets, metrics and related questions. The goal is to encourage a discussion about stimuli that can be useful to several groups at CBMM and have early feedback on the design of datasets and measurements to use.

CBMM Weekly Research Meeting

Nov 8, 2013 - 4:30 pm
Venue:  Harvard University: Northwest Bldg, Room 243 Address:  52 Oxford Street Harvard University Northwest Building Cambridge, 02138

There will be a presentation on the Research Thrust 5 “Theories for Intelligence.”
T. Poggio:

Thrust 5:  Theory is the common language between thrusts but it is not a goal in itself.  The CBMM challenge question what is there and m-theory

  • Theory: Invariances reduce sample complexity
  • Algorithm: how the ventral stream may learn/compute invariant features

Speakers:

  • Qianli Liao  – Computational test: algorithm performs well in face verification
  • Joel Leibo – Neuroscience test: algorithm is consistent with physiology of face cells
  • Leslie Valiant – The neuroidal model: modeling how neural circuits are cumulatively learned

Pages

Subscribe to Weekly Research Meetings