| |
| Optimization
and Generalization in Boosting
Ling Li, Yaser Abu-Mostafa
Abstract.
The superior out-of-sample performance of AdaBoost has been attributed
to the fact that it minimizes a cost function based on margin. In order
to examine how the cost function, in and of itself, affects the out-of-sample
performance, we apply several more sophisticated optimization techniques
directly to the cost function. When the AdaBoost exponential cost function
is optimized, our methods generally yield much lower cost and training
error but higher test error, which implies that the exponential cost
is vulnerable to overfitting. With the optimization power gained, we
can adopt more "regularized" cost functions that have better
out-of-sample performance but are difficult to optimize. Our experiments
demonstrate that with suitable cost functions, our methods can have
better out-of-sample performance. (full
report)
|
| Monotonic
Bernoulli Trials
Amrit Pratap, Yaser Abu-Mostafa, Pietro Perona
Abstract.
When estimating a number of bernoulli variables which have a certain
monotonicity constraint, if the number of samples for each variable
is small, then the estimates will not satisfy the monotonicity constraint.
Better performance is achieved by endorcing the monotonicity constraint
on the estimation procedure. (full
report)
|
| Path-Planning
for Feature-Recognition and Classification using Information Theoretic
Methods
Tim Chung, Joel Burdick, Richard Murray
Abstract.This
project investigates the role of information-theoretic techniques in
cooperative multi-agent systems. These techniques are used to govern
the path planning of agents to optimally classify features of interest
by improving the quality of the measurements. Sensor measurements are
assumed to be in the presence of noise. We consider issues associated
with distributed systems such as sensor fusion of information and formation
control of relative vehicle locations. The objective is to articulate
the theory underlying the relationship between sensing tasks and cooperative
control. (full report)
|
| Networks,
Evolution, Science & Neural Systems
Alex Bäcker
Abstract.
Recent times have seen the advent of large amounts of data on networks
of diverse kinds, from the WWW and citation networks to protein and
gene expression networks. Part of my work has been aimed at extracting
insight out of these massive collections of data. We show, for example,
that recent years have seen an expansion in the memory of science and
a homogenization of citation distributions. In parallel, I have been
developing mathematical methods to extract information from multi-neuron
recordings of brain activity. More generally, I am addressing a variety
of open questions at the interface of biology, math and computation.
(full report)
|
| Fast
Bayesian Support Vector Machine Parameter Tuning with the Nystrom Method
Carl Gold, Alex Holub
Abstract.
We experiment with speeding up a Bayesian method for tuning the
hyperparameters of a Support Vector Machine (SVM) classifier. The Bayesian
approach gives the gradients of the evidence as averages over the posterior,
which can be approximated using Hybrid Monte Carlo simulation (HMC).
By using the Nystrom approximation to the SVM kernel, our method significantly
reduces the dimensionality of the space to be simulated in the HMC.
We show that this speeds up the running time of the HMC simulation from
O(n^2) (with a large prefactor) to effectively O(n), where n is the
number of training samples. We conclude that the Nystrom approximation
has an almost insignificant effect on the performance of the algorithm
when compared to the full Bayesian method, and gives excellent performance
in comparison with other approaches to hyperparameter tuning.
|
| Computational
Modeling of Feature Inheritance
Whee Ky (Wei Ji) Ma
The proposition
that reentrant interactions into the early visual system are necessary
for visual awareness has lately been under close scrutiny. We examine
this proposition in the context of a neuronal model which explains the
phenomenology of feature inheritance. (full
report)
|
| Suppressive
Effect of Sustained Low-Contrast Adaptation followed by Transient High-Contrast
on Peripheral Target Detection
Farshad Moradi, Shinsuke Shimojo, Christof Koch
Filling-in
can be induced by high-contrast edge adaptation, or after prolonged
adaptation to a peripheral low-contrast object (Troxler fading). Adaptation
to sustained low-contrast vs. adaptation to transient high-contrast
suggests synergy between contrast and edge adaptation, but the possible
interactions are not well understood. We observed that briefly increasing
the contrast of a peripheral low-contrast object after a few seconds
of strict fixation elicits disappearance of the object, resulting in
perceptual filling-in of the location with the surround (Figure 1a).
After a short time usually around one second the object reappears. Hence,
following sustained adaptation to a low-contrast target, transient high-contrast
stimulation can induce perceptual disappearance. (full
report)
|
| Computational
Modeling of Visual Attention Systems
Robert J. Peters, Asha Iyer, Christof Koch
Nathan Mundhenk, Laurent Itti
Abstract.
We have continued to extend our biological model of bottom-up visual
attention with several recently characterized retinal and cortical interactions
that are known to govern human performance in certain visual tasks.
We are testing the behavioral importance of these interactions by comparing
our model's predictions against human eye movement data recorded with
our infrared eyetracker. In the last year we have worked with three
new model components: (1) short-range orientation interactions (for
clutter reduction), (2) long-range orientation interactions (for contour
facilitation), and (3) retinal filtering (for fovea vs. periphery effects).
(full report)
|
| Attentional
Selection for Learning and Recognition of Objects in Cluttered Scenes
Ueli Rutishauser, Dirk Walther, Christof Koch, and Pietro Perona
The problem
of serial processing of highly complex visual stimuli containing multiple
objects is not only faced by humans and other primates, but also by
machine vision systems. Advanced object recognition algorithms are capable
of achieving very good recognition performance with objects learned
from a single image (one-shot learning). These algorithms perform well
as long as they are trained on images in which a major part of the image
is occupied by the object to be learned and recognized. As soon as major
parts of an image are occupied by clutter it becomes impossible to learn
from such images without manual pre-labeling. These approaches are thus
not suitable in an unsupervised environment, as they would mainly learn
background clutter instead of the actual objects. (full
report)
|
| Automated
Event Detection in Underwater Video
Dirk Walther, Duane Edgington, Karen A. Salamy, Michael Risi, R. E. Sherlock,
and Christof Koch
Remotely
operated underwater vehicles (ROVs) become increasingly important as
a tool for obtaining quantitative data on the distribution and abundance
of oceanic animals. Using video cameras, it is possible to make quantitative
video transects (QVT) through the water, providing high-resolution data
at the scale of the individual animals and their natural aggregation
patterns. The current manual method of analyzing QVT video by trained
scientists is very labor intensive and poses a serious limitation to
the amount of data that can be obtained from ROV dives. (full
report)
|
| Attentional
Selection for Object Recognition in Visual Cortex
Dirk Walther, Laurent Itti, Maximilian Riesenhuber, Tomaso Poggio, Christof
Koch
Most models
of object recognition assume the isolated occurrence of objects in the
field of view. However, in our everyday experience we are usually confronted
with scenes that are cluttered with a variety of objects – some
relevant for our actions, some not. Our brain’s response to this
overwhelming flood of visual information is serializing the processing
of the objects by mechanisms of visual attention. Attentional selection
of objects is often modeled using all-or-nothing switching of neuronal
connection pathways from the attended region of the retinal input to
the recognition units. However, there is little physiological evidence
for such all-or-none modulation in early areas. We have developed a
combined model for spatial attention and object recognition in which
the recognition system monitors the entire visual field, but attentional
modulation by as little as 20% at a high level is sufficient to recognize
multiple objects. (full report)
|
| Evolutionary
Design Synthesis – From Sensors to Controllers
Yizhen Zhang, Alcherio Martinoli, Erik Antonsson
Collaborators: Jonathan Litt, Edmond Wong (NASA Glenn Center
Abstract.
In this project, an automated engineering design synthesis methodology
based on evolutionary methodology is being explored, with special interest
on design and optimization of distributed embodied systems. Two case
studies have been considered so far; the first one concerns the design
of a collective sensory system for traffic monitoring purposes, while
the second one deals with the development of neural-based robot controllers
for turbine blades inspection. It has been shown that the evolutionary
methodology is able to address the engineering design challenges present
in the case studies as well as other complex design problems, and synthesize
novel design solutions of good quality. Moreover, the fitness function
can be formulated as an aggregation of fuzzy design preferences with
different weights and trade-off strategies leading to an automatic generation
of the complete Pareto-optimal frontier.. (full
report)
|
| Spike
Based Saliency Detection
Ulrik Beierholm, Pietro Perona
Abstract.
Trying to quickly ascertain which parts of a visual scene is most relevant
for a recognition task and then focusing on each of these areas, is
an economical use of processing power known to be employed in the human
visual system. Most models for saliency detection however are too slow
to explain the performance of the biological system. We are currently
working on implementing a fast neuronal spike based saliency detector
model based on rank order coding. (full
report)
|
| Decomposition
of Human Motion into Dynamics Based
Primitives with Application to Drawing Tasks
Domatilla Del Vecchio, Richard Murray, Pietro Perona
Abstract.
Using tools from dynamical systems and systems identification we
develop a framework for the study of primitives for human motion, which
we refer to as movemes. The objective is understanding human motion
by decomposing it into a sequence of elementary building blocks that
belong to a known alphabet of dynamical systems. We develop a segmentation
and classification algorithm in order to reduce a complex activity into
the sequence of movemes that have generated it. We test our ideas on
data sampled from five human subjects who were drawing figures using
a computer mouse. Our experiments show that we are able to distinguish
between movemes and recognize them even when they take place in activities
containing an unspecified number of movemes. (full
report)
|
| Human
Motion Detection and Classification
Claudio Fanti, Pietro Perona
Abstract.
We foresee a future in which machines autonomously interact with Humans
in the surrounding environment. So far, very good results have been
achieved in detecting the presence of Humans and labeling their body
parts by means of graphical-models based algorithms. We unavoidably
have to deal with uncertainty and reasoning in absence of complete information.
To that extent, we explore and enhance the state of the art in probabilistic
inference and sampling techniques having the machines understanding
human actions as a primary application. (full
report)
|
| Dynamic
Recurrent Neural Networks for Pattern Recognition
Alex Holub, Gilles Laurent, Pietro Perona
We are
investigating the computational properties of recurrent neural networks
of binary artificial neurons. Our investigations are guided by recent
work performed in the laboratory of Gilles Laurent which involves elucidating
the underlying processing mechanisms in early olfactory processing.
These physiological investigations indicate that the initial olfactory
processing layer (in the locust the antennal lobe) consists of a dynamic
recurrent neural network of excitatory and inhibitory units. The presentation
of stimuli to the network results in stereotyped spatio-temporal neural
firing patterns, with each unique stimulus presentation invoking a unique
temporally-varying pattern of activity within the population of neurons.
We have approximated the biological networks using recurrent networks
with discrete binary neural elements. These non-linear networks exhibit
chaotic behavior such that similar input patterns obtain very dissimilar
network representations through the network dynamics. Similar pattern
spreading characteristics have been observed in the initial processing
networks of fish by members of the Laurent laboratory and it has been
hypothesized that pattern spreading may be one computational benefit
which the initial processing layer provides. (full
report)
|
| Rapid
Natural Scene Categorization without Attention
Fei Fei Li, Rufin VanRullen, Christof Koch, Pietro Perona
Abstract.
What can we see when we do not pay attention? While attention is not
necessary for some detection tasks on simple synthetic stimuli, without
it we are “blind” even to major aspects of a natural complex
scene. It would thus appear that only visual tasks that have an explanation
in the early stages of the visual system may be carried out without
attention. We report on a complex visual task that requires no attention.
Our subjects can rapidly detect animals in briefly presented natural
scenes while simultaneously performing another visual task that demands
full attention. By comparison, they are unable to discriminate large
‘T’s from ‘L’s in the same conditions. We conclude
that attention may not be necessary for some visual tasks that are associated
with ‘high level’ cortical areas. (full
report)
|
| Object
Categorization: Unsupervised One-Shot Learning
Fei-Fei Li, Rob Fergus, Pietro Perona
Abstract.
Learning visual models of object categories notoriously requires thousands
of training examples; this is due to the diversity and richness of object
appearance which requires models containing hundreds of parameters.
We present a method for learning object categories from just a few images
(1 - 5). It is based on incorporating "generic'' knowledge which
may be obtained from previously learnt models of unrelated categories.
We operate in a variational Bayesian framework: object categories are
represented by probabilistic models, and "prior'' knowledge is
represented as a probability density function on the parameters of these
models. The "posterior'' model for an object category is obtained
by updating the prior in the light of one or more observations. Our
ideas are demonstrated on four diverse categories (human faces, airplanes,
motorcycles, spotted cats). Initially three categories are learnt from
hundreds of training examples, and a "prior'' is estimated from
these. Then the model of the fourth category is learnt from 1 to 5 training
examples, and is used for detecting new exemplars a set of test images.
(full report)
|
| Object
Recognition by Probabilistic Hypothesis Construction
Pierre Moreels, Michael Maire, Pietro Perona
Abstract.
We present a probabilistic framework for recognizing objects in
images of cluttered scenes. Hundreds of objects may be considered and
searched in parallel. Each object is learned from a single training
image and is modeled by the visual appearance of a set of features,
as well as their position with respect to a common reference frame.
The recognition process computes both the identity and position of objects
in the scene by computing the best interpretation (or hypothesis) of
the scene in the light of a database of known objects. A hypothesis
pairs features in an input image either with features in the database
or marks them as clutters. Each hypothesis may be scored in a principled
way using a generative model of the image which is defined using the
learned objects as well as a model for clutter. While the space of all
possible hypotheses is enormously large, one may find the best hypothesis
efficiently—we explore a couple of heuristics to do so. In our
initial experiments our algorithm compares favorably with state-of-the-art
recognition systems. (full
report)
|
| Perception
of Mirror Surfaces
Silvio Savarese, Fei Fei Li, Pietro Perona
Abstract.
The aim of our work is to investigate how the human visual system perceives
specular surfaces and which cues can be used to recover the shape of
such class of objects. Our experiments show that mirror reflections
are a weak cue for most human observers when additional information
is not available. (full report)
|
| 3D
Reconstruction of Specular Surfaces
Silvio Savarese, Min Chen and Pietro Perona
Abstract.
Specular reflections carry valuable information on surface shapes.
A curved mirror surface produces "distorted" images of the
surrounding world. For example a straight line reflected by a curved
mirror is in general a curve. It is clear that such distortions are
systematically related to the shape of the surface. Our goal is to explore
the geometry linking the shape of a curved mirror surface to the distortions
produced on a scene it reflects. To this effect, we assume a simple
known (calibrated) scene composed of lines passing through a point.
We demonstrate that local shape geometry of the surface may be recovered
from local deformation of the reflected images of at least three intersecting
lines. (full report)
|
|
