This page contains a list
of some of the papers I have published over the last few years:
Authors: Nada Matic, John Platt, Tony Wang
Download
this paper: (PDF)
This
paper presents QuickStroke: a system for the incremental recognition of
handwritten Chinese characters. Only a few strokes of an ideogram need to be
entered in order for a character to be successfully recognized. Incremental
recognition is a new approach for on-line recognition of ideographic
characters. It allows a user to enter characters a factor of 2 times faster
than systems that require entry of full characters. Incremental recognition is
performed by a two-stage system which utilizes 68 neural networks with more
than 5 million free parameters. To enable incremental recognition, we use
specialized time-delay neural networks (TDNNs) that are trained to recognize partial
characters. To boost the recognition accuracy of complete characters, we also
use standard fully-connected neural networks. Quickstroke is 97.3% accurate for
the incremental writer-independent recognition of 4400 simplified GB Chinese
ideograms.
Reference: International
Conference on Pattern Recognition, 2002, to appear.
Authors: John Platt and Nada Matic
Download this paper: (PDF) (DjVu) (DjVu plug-in)
This paper discusses a
fairly general adaptation algorithm which augments a standard neural network to
increase its recognition accuracy for a specific user. The basis for the
algorithm is that the output of a neural network is characteristic of the
input, even when the output is incorrect. We exploit this characteristic output
by using an Output Adaptation Module (OAM) which maps this output into
the correct user-dependent confidence vector. The OAM is a simplified Resource
Allocating Network which constructs radial basis functions on-line. We applied
the OAM to construct a writer-adaptive character recognition system for on-line
hand-printed characters. The OAM decreases the word error rate on a test set by
an average of 45%, while creating only 3 to 25 basis functions for each writer
in the test set.
Reference: Advances in Neural Information
Processing Systems 9, M. Mozer, M. Jordan, T. Petsche, eds., pp. 765-771, MIT
Press, (1997).
Authors: John Platt and Tim Allen
Download this paper: (PDF) (DjVu) (DjVu plug-in)
This paper describes a
neural network classifier for the I1000 chip, which optically reads the E13B font
characters at the bottom of checks. The first layer of the neural network is a
hardware linear classifier which recognizes the characters in this font. A
second software neural layer is implemented on an inexpensive microprocessor to
clean up the results of the first layer. The hardware linear classifier is
mathematically specified using constraints and an optimization principle. The
weights of the classifier are found using the active set method, similar to
Vapnik's separating hyperplane algorithm. In 7.5 minutes of SPARC 2 time, the
method solves for 1523 Lagrange multipliers, which is equivalent to training on
a data set of approximately 128,000 examples. The resulting network performs
quite well: when tested on a test set of 1500 real checks, it has a 99.995%
character accuracy rate.
Reference: Advances in Neural Information
Processing Systems 8, D. Touretzky, M. Mozer, M. Hasselmo, eds., pp. 938-944,
MIT Press, (1996).
Authors: Steve Nowlan and John Platt
Download this paper: (Postscript) (DjVu) (DjVu plug-in)
We describe a system
which can track a hand in a sequence of video frames and recognize hand
gestures in a user-independent manner. The system locates the hand in each
video frame and determines if the hand is open or closed. The tracking system
is able to track the hand to within 10 pixels of its correct location in 99.7%
of the frames from a test set containing video sequences from 18 different
individuals captured in 18 different room environments. The gesture recognition
network correctly determines if the hand being tracked is open or closed in
99.1% of the frames in this test set. The system has been designed to operate
in real time with existing hardware.
Reference: Advances in Neural Information
Processing 7, G. Tesauro, D. Touretzky, T. Leen, eds., pp. 901-908, MIT Press,
(1995).
Authors: Ralph Wolf and John Platt
Download this paper: (PDF) (DjVu) (DjVu plug-in)
This paper describes the
use of a convolutional neural network to perform address block location on
machine-printed mail pieces. Locating the address block is a difficult object
recognition problem because there is often a large amount of extraneous
printing on a mail piece and because address blocks vary dramatically in size
and shape.
We used a convolutional
locator network with four outputs, each trained to find a different corner of
the address block. A simple set of rules was used to generate ABL candidates
from the network output. The system performs very well: when allowed five
guesses, the network will tightly bound the address delivery information in
98.2% of the cases.
Reference: Advances in Neural Information
Processing Systems 6, J. Cowan, G. Tesauro, J. Alspector, eds., pp. 745-752,
Morgan-Kaufmann, (1994).
Author:
Abstract:
This paper presents a
constraints method for physically based computer graphics models, based on the
constraint stabilization method of Baumgarte and on the dynamic constraints of
Barzel and Barr. These new constraints are called generalized dynamic
constraints (GDCs). GDCs extend dynamic constraints to obey the principle of
virtual work and to fulfill time-varying and inequality constraints. The
constraint forces of GDCs are computed by a sparse linear system and are
proportional to the Lagrange multipliers of the constraints. GDCs are used to
assemble deformable computer graphics models and to simulate collisions between
the models.
Reference: CVGIP: Graphical Models and Image
Processing, vol. 54, no. 6, November, pp. 516-525, (1992).
Author: John Platt
Download this paper (PDF)
We have created a network
that allocates a new computational unit whenever an unusual pattern is
presented to the network. This network forms compact representations, yet learns
easily and rapidly. The network can be used at any time in the learning process
and the learning patterns do not have to be repeated. The units in this network
respond to only a local region of the space of input values.
The network learns by
allocating new units and adjusting the parameters of existing units. If the
network performs poorly on a presented pattern, then a new unit is allocated
which corrects the response to the presented pattern. If the network performs
well on a presented pattern, then the network parameters are updated using
standard LMS gradient descent.
We have obtained good
results with our resource-allocating network (RAN). For predicting the Mackey
Glass chaotic time series, our network learns much faster than do those using back-propagation
and uses a comparable number of synapses.
Reference: Neural Computation, vol. 3, no. 2, pp.
213-225, (Summer 1991). Copyright 1991 by the Massachusetts Institute of
Technology.
Authors: John Platt and Federico Faggin
Download this paper: (PDF) (DjVu) (DjVu plug-in)
We have created new
networks to unmix signals which have been mixed either with time delays or via
filtering. We first show that a subset of the Herault-Jutten learning rules
fulfills a principle of minimum output power. We then apply this principle to
extensions of the Herault-Jutten network which have delays in the feedback
path. Our networks perform well on real speech and music signals that have been
mixed using time delays or filtering.
Reference: Advances in Neural Information
Processing 4, J. Moody, S. Hanson, R. Lippmann, eds., pp. 730-737,
Morgan-Kaufmann, (1992).
This page was written by John Platt in the Knowledge Tools Group of Microsoft Research. Last updated: 04/20/2006