[ont.events] U of Toronto Computer Science activities for June 20-30

clarke@csri.toronto.edu (Jim Clarke) (06/17/88)

         (SF = Sandford Fleming Building, 10 King's College Road)
              (GB = Galbraith Building, 35 St. George Street)

SUMMARY:

GRAPHICS AND INTERACTION SEMINAR, Tuesday, June 21, 3 pm, GB120 -- Paul Borrel:
     "Interactive Design with Sequences of Parameterized Transformations"

THEORY SEMINAR, Monday, June 27, 11 am, GB244 -- Joan Boyar:

A.I. SEMINAR, Monday, June 27, 3 pm, SF1101 -- Allen M. Waxman:
     "CONVECTED ACTIVATION PROFILES: Receptive Fields for Real-Time
          Measurement of Short-Range Visual Motion"

---------------

    GRAPHICS AND INTERACTION SEMINAR, Tuesday, June 21, 3:00 pm, GB120

                                Paul Borrel
Laboratoire d'Automatique et de Microe'lectronique de Montpellier - France
           Currently on Sabbatical at IBM Watson Research Center

  ``Interactive Design with Sequences of Parameterized Transformations''

We present a paradigm for capturing the functional and relational aspects
of a design  as a sequence  of parameterized unevaluated operations, which
may, for  example, correspond to the individual  steps of a manufacturing
process.  A sequence that can transform a wide variety of models in a
manner consistent  with the user's intentions  can be specified interac-
tively using a  single model as an  example. The execution of  a  sequence
transform  a  particular  model produces a result  which may be interro-
gated  to detect constraint violations and to  locate operations responsi-
ble for creating or modifying specific features. An experimental implemen-
tation in  an  object oriented  environement  is  described.

             THEORY SEMINAR, Monday, June 27, 11:00 am, GB244

                                Joan Boyar
                           University of Chicago

We define a new structured and general model of computation: circuits
using  arbitrary  fan-in  arithmetic  gates  over  the characteristic-two
finite fields. These circuits  have  only  one input  and  one  output. We
show how they correspond naturally to boolean computations with n inputs
and n outputs.  These circuits are  equivalent to threshold circuits in
that a simulation in ei- ther direction requires only a constant blow up in
depth  and  a polynomial blow up in size.

The threshold circuits we consider allow  arbitrary  integer weights.  How-
ever,  we  show that this is equivalent to the usual model of threshold
circuits, which is equivalent to majority cir- cuits  in  which  any  two
inputs to the same gate must come from different sources.

This is joint work with Gudmund Frandsen and Carl Sturtivant.

                A.I. SEMINAR, Monday, June 27, 3 pm, SF1101

                         Professor Allen M. Waxman
                      Laboratory for Sensory Robotics
                             Boston University
                                    and
                          MIT Lincoln Laboratory
                        Machine Intelligence Group

CONVECTED ACTIVATION PROFILES: Receptive Fields for Real-Time Measurement of
                         Short-Range Visual Motion

A new method is developed for the measurement of short-range visual motion
in image sequences by exploiting the motion of image features such as edges
and points. Each feature generates a Gaussian activation profile (or influ-
ence function) in a spatio-temporal neighborhood of specified scale around
the feature itself; this profile is then convected with the motion of the
feature.  Image featural velocity estimates may then be obtained from such
dynamic activation profiles using a modification of familiar gradient tech-
niques. The resulting estimators are formulated in terms of simple ratios
of spatio- temporal filters (i.e., receptive fields) convolved with image
feature map sequences. A family of activation profiles of varying scale
must be utilized to cover a range of image velocities. In order to decide
which velocity estimate is to be accepted, the different motion channels
undergo a characteristic speed normalization before they compete. This nor-
malization scheme is consistenet with human motion perception. This mul-
tilayered processing scheme is well suited for parallel implementation. I
will show a videotape which illustrates real-time motion estimation on the
PIPE machine with feature velocity maps generated at 15 updates per second.
-- 
Jim Clarke -- Dept. of Computer Science, Univ. of Toronto, Canada M5S 1A4
              (416) 978-4058
BITNET,CSNET: clarke@csri.toronto.edu     CDNNET: clarke@csri.toronto.cdn
UUCP: {allegra,cornell,decvax,linus,utzoo}!utcsri!clarke