B.A. in Computer Science from the Israel Institute of Technology Technion, S.M. in Electrical Engineering and Computer Science and Ph.D. in EECS from MIT
Research Scientist, NEC Research Institute, Inc.
Artificial Language, Cognitive Science, Machine Vision, Computational Linguistics, Computer Systems (including Visual Event Perception, Computational Models of Child Language Acquisition) and High Performance Compilation Strategies for LISP.
Founder, MetaLogic, Inc.
1986 One of 25 annual nationwide recipients of a four year AT&T Ph.D. fellowship, 1986–1990.
1992 Nominated for the ACM Distinguished Dissertation Award.
1993 The George M. Sprowls Award for ‘Outstanding contribution in the field of Electronic Computer and Investigation
Research by an EECS Student,’ Department of Electrical Engineering and Computer Science, MIT.
1996 The Louis and Miriam Benjamin Academic Lectureship, Department of Electrical Engineering, Technion.
2013 Best paper award, 51st Annual Meeting of the Association for Computational Linguistics (ACL).
2015 Paper published in senior member track, blue-sky ideas subtrack, Conference on Artificial Intelligence (AAAI).
Paper published in award winning papers track, Journal of Artificial Intelligence Research (JAIR).
IEEE, Computer Society.
Salient Boundary Detection Using Ration Contour, IEEE Transactions on Pattern Analysis and Machine Intelligence (2004). Reconstructing Force-Dynamic Models from Video Sequences, Artificial Intelligence (AIJ), 151 (December 2003).Publications include numerous book chapters, journal articles, conference papers, workshops and patents.
Prof. Siskind has redesigned the curriculum for ECE47300, an undergraduate course in AI. The traditional AI curriculum, which focusses on search, automated reasoning, and planning, is less relevant for today’s ECE undergraduates.
AI is primarily a research enterprise. Most undergrads find training in AI of little relevance to their anticipated career
in industry. Thus, Prof. Siskind has redesigned the curriculum for ECE47300 to focus on material that is relevant
to preparing ECE undergrads for an industrial career. This includes styles of programming that are not covered in
other ECE courses: functional programming and symbolic manipulation. Prof. Siskind has chosen to retain the
focus on functional programming and symbolic manipulation, and not refocus the course on current fads like deep
learning, because these fundamental skills can be applied much more broadly across all of computer engineering and
have stood the test of time. Since the computer engineering curriculum at Purdue is focussed primarily on hardware
design, students otherwise lack adequate preparation in software and algorithm design. Retaining the traditional focus
on functional programming and symbolic manipulation affords the opportunity to give students the sorely needed opportunity to expand their programming experiences with programming styles and algorithmic content to which they
would otherise not be exposed to. This provides lasting career-long value that would not be provided by teaching
students to use canned tools to train a classifier with backpropagation.
In this redesigned course, many techniques and algorithms from AI are taught within the context of solving ECE
problems, instead of traditional AI problems. For example, the concept of evaluation is taught by having the students
write an evaluator for Boolean expressions, rather than an evaluator for LISP. The concept of rewrite systems is
taught by having the students write a simplifier for Boolean expressions, rather than an expert system. The concept of
resolution is taught by having the students write a system that uses resolution to find faults in a digital circuit rather
than to prove theorems.
The redesigned course focuses on algorithms: evaluation, pattern matching and rewrite systems, constraint satisfaction,
and automated reasoning techniques like semantic tableaux, resolution, and congruence closure. It is difficult for
students to become fluent in these algorithms solely from the lectures. Thus, the problem sets have the students
implement most of the algorithms taught in lecture.
Significant effort has gone into preparing the problem sets. It is not feasible for students to implement the material
taught in lecture without a prepared infrastructure. There would simply be too much code to write over the course
of a semester. Thus, Prof. Siskind has prepared a framework for each problem set that includes three components:
(a) the low-level data-structure manipulation routines, (b) a search engine, and (c) a GUI that handles I/O and often
animates the algorithm in operation. Within this framework, students need only implement the concept-rich portion of
material taught in class. Having students write code that interoperates with a larger system teaches them how to read
and understand APIs and write code that conforms to specifications. Prof. Siskind makes this course material available
to other instructors, including ones at Georgia Tech and the University of Washington.
In 2010, Prof. Siskind redesigned the curriculum for ECE57000, a graduate course in AI. As part of the revised course
requirements, students do a term paper/project/presentation. Students select and read three papers published within the
past three years in a conference or journal in AI, broadly construed to include AI, computer vision, natural-language
processing, robotics, machine learning, cognitive science, and neuroscience. They then implement the ideas in one of
them and write a six-page term paper in AAAI submission format, three pages of which present a review and critique of
the three papers that they read and three pages of which describe their implementation and the experiments/evaluation
that they performed. Upon completion, they present a 12- to 25-minute conference-style PowerPoint presentation to
the class, half of which presents a review and critique of the three papers that they read and half of which discusses
their implementation and evaluation. Inter alia, this has been used to satisfy the CS department Communication Requirement for MS students posted at https://www.cs.purdue.edu/graduate/curriculum/masters.
In 2016 and 2017, Prof. Siskind further redesigned the curriculum for ECE57000 to focus on machine learning in
general and deep learning in particular. The lectures contain newly developed material that teaches the fundamentals of forward and reverse mode automatic differentiation, teaches back propagation and training of neural networks
using automatic differentiation, and teaches the fundamentals of object detection in computer vision using convolutional neural networks. Newly designed problem sets for this course have students learn to use multiple deep learning
frameworks, including TORCH and CAFFE, in multiple programming languages, including LUA and PYTHON, and
build complete end-to-end systems to do classification using multilayer perceptrons, object classification using convolutional neural networks, and object detection and localization using proposal-general mechanisms together with
object classifiers. This is in addition to the term paper/project/presentation discussed above.