STATUS
I am a graduate student at MIT, working with Sebastian Seung. I completed my undergraduate education in 2004, at the University of Pennsylvania.
INTERESTS I am interested in two
main questions:
1. How does network structure influence
computational function in neural systems? 2. How can complex
systems use experience to improve performance?
Various
forms of these questions are today pursued in many different fields:
neuroscience, machine learning, robotics, cognitive science, statistics,
control theory, etc. Perhaps most exciting is the prospect of common principles which
underlie all of these different points of view, and the application of such principles to the creation of new technology.
PROJECTS
1. Computational Neuroanatomy. Nervous systems are
circuits. Measurements of neural activity have observed that
information in the brain flows in specific spatiotemporal patterns.
Beyond external stimuli and intrinsic cellular properties, what determines these
patterns of activity? A likely candidate is the way in which neural
systems are wired together, i.e. their circuit diagram. In
principle, recovering such circuit diagrams is not hard: simply trace
the wires that connect brain cells to each other
and store a list of the connections. In practice, this is
incredibly difficult. For example, axons can be microscopic in size
(50-100 nanometers) while macroscopic in extent (centimeters).
New imaging technologies with sufficient resolution,
field of view, and automation are beginning to provide the raw measurements
necessary for reconstructing circuit diagrams. However, the data that
such imaging produces is so vast in quantity and complex in form that
sophisticated computational tools are necessary for their use. We
are developing these tools, which at this stage largely involves
research at the intersection of machine learning and computer vision
(see [4]).
2. Neural network architectures for perception. Perceptual
systems have been heavily studied in both neurobiology and computer
science. I am particularly interested in how certain properties observed
in biological visual systems (such as structural hierarchies, feedback
loops, local receptive fields, and plasticity) can be understood in
computational terms, and how such insights can guide the construction of
more effective computer vision systems. More specifically, this involves
the theoretical analysis between network structure and function (see [3]) as well as the construction of artificial systems that
can learn from data. (see [4]).
PAPERS
[4] Supervised Learning of Image Restoration
with Convolutional Networks. Viren Jain, Joseph F. Murray,
Fabian Roth, Srinivas Turaga, Valentin Zhigulin, Kevin Briggman, Moritz
Helmstaedter, Winfried Denk, and H. Sebastian Seung. International
Conference on Computer Vision [ICCV 2007] [paper] [supplementary: specific
methods, video: 4 MB small, 16 MB large]
[3]Representing Part-Whole Relations in
Recurrent Neural Networks. Viren Jain, Valentin Zhigulin, and
H. Sebastian Seung. Advances in Neural Information Processing
Systems 18 [NIPS 2006] [paper]
[2]Exploratory analysis and visualization
of speech and music by locally linear embedding. Viren Jain and
Lawrence K. Saul. International Conf. on Speech, Acoustics, & Signal
Processing [ICASSP 2004] [paper]
[1]A Smorgasbord of Features for
Statistical Machine Translation. Franz Och, Daniel Gildea,
Sanjeev Khudanpur, Anoop Sarkar, Kenji Yamada, Alex Fraser, Shankar
Kumar, Libin Shen, David Smith, Katherine Eng, Viren Jain, Zhen Jin, and
Dragomir Radev. Human Language Technology and 5th Meeting of the
NAACL [HLT-NAACL 2004] [paper]
