V1 hypercolumn Coordination Meeting, 20th - 21st Nov 2006

Comments

Source

Subject: Coordination meeting of the WP9T2 and WP5T3 tasks.

  • The goal of the meeting was to prepare the next deliverable D25 ("model of an hyper-column") but also to join our efforts in modeling. In particular, important decisions were made toward finding canonical parameters (structure, neural parameters) of all systems being delivered in WP5T3 and WP9T2 but also concerning the definition of the common benchmark that will de deployed to the different implementations in the different partners. From this benchmark (benchmark zero), we should be able to validate different solutions to pinpoint their strengths and weaknesses.

Main points and decisions

  • review / demonstration of coding strategies by different partners. This definition has impact on the "Meta Simulator",
  • review of existing V1 models inside and outside FACETS (see also !Review_of_V1_models),
  • definition of the common benchmark FACETS_Benchmarks to qualitatively compare different models and strategies in order to break the model complexity barrier, as we began to do for the simulator,
  • definition of the architecture of the FACETS model of an "hyper-column" by defining the priorities of different partners: effect of including different neuron types, including layers, specific lateral connectivity.
  • Partner 13 (Plymouth, Mike Denham) will provide an hypercolumn for D25.
  • Partner 6b will provide a retina before Dec. 15 for implementing benchmark zero.

Program

(actual program changed slightly)

Monday, 20 November 2006

Monday, 20 November 2006

morning, 09:00 - 13:00

Panel discussion : Reviewing essential knowledge from biology

* Christophe Lamy - Alex Thomson (ULON) : Diversity in the architecture
(layering, cell types and dynamics) of cortical columns
* Fred Chavane (INCM) : Essentials of the functional architecture
and cortical lateral connectivity
* Cyril Monier (UNIC) : Models of V1 and orientation selectivity
(also see
Review_of_V1_models)
Round-table : what shall we keep for modeling/hardware ? what are the
priorities ? Moderator: A. Lansner

13:00

Lunch

afternoon

Visual benchmarking for FACETS V1 models

* Laurent Perrinet
(INCM) : Short description of visual function benchmarks for FACETS
* Jens Kremkow (INCM): Contrast gain control with layered neural
network
* Guillaume Masson (INCM): Using IO data for tuning models ?
* Adrien Wohrer (INRIA) : (CANCELED) Statistics of inputs for
benchmarking V1 models)Round-table : Implementation plans for
benchmarking. Moderator : O. Faugeras

Tuesday, 21 November 2006

Tuesday, 21 November 2006

morning, 09:00 - 13:00

Overview of V1 models from FACETS: diversity and common strengths

* Mike Denham (UP) : The english model
* Anders Lansner (KTH) : The swedish model
* Malte Rasch / Wolfgang Maass (TUG) : The austrian model
* Outcome : Definition of a FACETS common model of V1. Moderator : L.
Perrinet

13:00

Lunch

afternoon

Planning of Implementation plans * benchmarking the existing
models
* implementing the FACETS model for D25
* link between WP9 and WP5 on the track to the "detailed model" of
V1 computing.
Outcome : to set deadlines and implementation plans. Moderator : G.
Masson

venue

who, when

  • complementary to the FACETS Web Site here are the timetable for participants (opportunities to share a taxi, there?):
  • Olivier Faugeras will attend on monday (no hotel)
  • Andrew Davison will arrive on Tuesday, leaves wednesday afternoon (hotel on tuesday)
  • Cyril Monier (hotel from sunday to wednesday)
  • Olivier Marre Pierre YGER will arrive on Sunday, leave wednesday morning (hotel from Sunday to Tuesday)
  • Anders Lansner + Martin Rehn arrive Sunday evening and leave Tuesday (flight around 5 pm). (hotel from sunday to monday)
  • Malte Rasch will arrive on Sunday 19.11 (hotel from sunday to monday)
  • Christophe Lamy will arrive on Sunday @ 11.30pm leave monday (hotel on sunday)
  • Michael Denham arives monday at 2pm (st charles), leaves tuesday at 18.09 (st charles) - (hotel on monday)

Comments

Comments powered by Disqus