VS298: Unsolved Problems in Vision

From RedwoodCenter
Revision as of 00:41, 30 August 2014 by Bruno (talk | contribs)
Jump to navigationJump to search

One of the goals of vision science is to understand the nature of perception and its neural substrates. There are now many well established techniques and paradigms in both psychophysics and neuroscience to address problems in vision. However, knowing how to frame these questions for investigation is not necessarily obvious. Nervous systems present us with stunning complexity, and the purpose of perception itself is deeply mysterious. The goal of this seminar course is to step back and ask, what are the important problems that remain unsolved in vision research, and how should these be approached empirically? The course will consist of alternating weeks of discussion and guest lectures by vision scientists who will frame their views of the core unsolved problems. Interdisciplinary groups of students will devise a practical research plan to address an unsolved problem of their choice.

Instructors: Stan Klein, Jerry Feldman, Bruno Olshausen, and Karl Zipser
GSI: [1]

Enrollment information:

VS 298 (section 2), 2 units
CCN: 66478

Meeting time and place:

Tuesday 6-8, 489 Minor

Email list:

vs298-unsolved-problems@lists.berkeley.edu subscribe


Weekly schedule:

Date Topic/Reading Presenter
Feb. 3 Redundancy reduction, whitening, and power spectrum of natural images
  • Barlow (1961): Theory of redundancy reduction paper
  • Atick (1992): Theory of whitening paper
  • Field (1987): 1/f2 power spectrum and sparse coding paper

Additional reading:

  • Attneave (1954) - 'Some informational aspects of visual perception' paper
  • Laughlin (1981) - Histogram equalization of contrast response paper
  • Srinivasan (1982) - 'Predictive coding: a fresh view of inhibition in the retina' paper
  • Switkes (1978) - Power spectrum of carpentered environments paper
  • Ruderman (1997) - Why are images 1/f2? paper
  • Torralba & Oliva (2003) - Power spectrum of natural image categories paper

Anthony DiFranco
Dylan Paiton
Michael Levy

Feb. 10 Whitening in time and color; Robust coding
  • Dong & Atick (1995): spatiotemporal power spectrum of natural movies paper
  • Ruderman (1998): statistics of cone responses paper
  • Karklin & Simoncelli (2012): noisy population coding of natural images paper

Additional reading:

  • Dong & Atick (1995) - spatiotemporal decorrelation using lagged and non-lagged cells paper
  • Doi & Lewicki (2007) - A theory of retinal population coding paper

Chayut Thanapirom
Michael Levy
Yubei Chen

Feb. 17 ** Holiday **
Feb. 24 Higher-order statistics and sensory coding
  • Barlow (1972): Sparse coding paper
  • Field (1994): What is the goal of sensory coding? paper
  • Bell & Sejnowski (1995): Independent component analysis. paper

Additional reading:

  • Redlich (1993): Redundancy Reduction as a Strategy for Unsupervised Learning. paper
  • Baddeley (1996): Searching for filter with 'interesting' output distributions: An uninteresting direction to explore? paper
  • O'regan & Noe (2001): A sensorimotor account of vision and visual consciousness paper

Karl Zipser
Michael Levy
Mayur Mudigonda

March 3 ICA and sparse coding of natural images
  • Bell & Sejnowski (1997): ICA of natural images paper
  • Olshausen & Field (1997): Sparse coding of natural images paper
  • van Hateren & Ruderman (1998), Olshausen (2003): ICA/sparse coding of natural video paper1, paper2

Additional reading:

  • Olshausen & Field (1996): simpler explanation of sparse coding paper

Mayur Mudigonda
Zayd Enam
Georgios Exarchakis

March 11 **Tuesday** Statistics of natural sound and auditory coding
  • Clark & Voss: '1/f noise and music' paper
  • Smith & Lewicki: sparse coding of natural sound paper
  • Klein/Deweese: ICA/sparse coding of spectrograms paper1, paper2

Tyler Lee
Yubei Chen
TBD

March 17 Higher-order group structure
  • Geisler: contour statistics paper
  • Hyvarinen: subspace ICA/topgraphic ICA paper1, paper2
  • Lyu & Simoncelli: radial Gaussianization paper

Additional reading:

  • Parent & Zucker (1989): Trace Inference, Curvature Consistency, and Curve Detection, paper
  • Field et al. (1993): Contour Integration by the Human Visual System: Evidence for a Local “Association Field” paper
  • Zetzsche et al. (1999): The atoms of vision: Cartesian or polar? paper
  • Garrigues & Olshausen (2010): Group Sparse Coding with a Laplacian Scale Mixture Prior, paper

Chayut Thanapirom
Guy Isely
TBD

March 24 ** Spring recess **
March 31 Energy-based models
  • Hinton: Product of experts models, paper
  • Osindero & Hinton: Product of Experts model of natural images, paper
  • Roth & Black: Fields of experts, paper

Additional reading:

  • Hinton: Practical guide to training RBMs paper
  • Teh et al: Energy-based models for sparse overcomplete representation, paper
  • Zhu, Wu & Mumford: FRAME (Filters, random fields, and maximum entropy), paper

Evan Shelhamer
Brian Cheung
Chris Warner

April 7 Learning invariances through 'slow feature analysis'
  • Foldiak/Wiskott: slow feature analysis, paper1, paper2
  • Hyvarinen: 'Bubbles' paper
  • Berkes et al.: factorizing 'what' and 'where' from video, paper

Guy Isely
Chayut Thanapirom
Bharath Hariharan

April 14 Manifold and Lie group models
  • Carlsson et al.: Klein bottle model of natural images, paper
  • Culpepper & Olshausen: Learning manifold transport operators, paper
  • Roweis & Saul: Local Linear Embedding, paper

Yubei Chen
Bruno/Mayur
James Arnemann

April 21 Hierarchical models
  • Karklin & Lewicki (2003): density components, paper
  • Shan & Cottrell: stacked ICA, paper
  • Cadieu & Olshausen (2012): learning intermediate representations of form and motion, paper

Tyler Lee
Brian Cheung
Dylan Paiton

April 28 Deep network models
  • Hinton & Salakhudinov (2006): stacked RBMs, paper
  • Le et al. (2011): Unsupervised learning (Google brain, 'cat' neurons), paper
  • Krishevsky et al. (2012): Supervised learning, ImageNet 1000 paper

TBD
TBD
Reza Abbasi-Asl

May 6

Note: Tuesday

Special topics
  • Fergus (2013): visualizing what deep nets learn paper
  • Schmidhuber: deep nets (paper), focusing on LOCOCODE (paper)
  • Image compression with Hopfield networks

Shiry Ginosar
Anthony DiFranco
Chris Hillar

May 12 Special topics