June
10, 2025
Mitigating
Biases in Self-consuming Generative Models
5:30 PM - 6:30 PM PT
Speaker:
Dr. Ali Siakoohi
Register at: https://events.vtools.ieee.org/m/486945
In this
talk, Dr. Ali Siahkoohi highlights the risks
of the current industrial AI practices
involving training large-scale generative
models on vast amounts of data scraped from
the internet. This process unwittingly leads
to training newer models on increasing amounts
of AI-synthesized data that is rapidly
proliferating online, a phenomenon Dr.
Siahkoohi refers to as ``model autophagy''
(self-consuming models). He shows that without
a sufficient influx of fresh, real data at
each stage of an autophagous loop, future
generative models will inevitably suffer a
decline in either quality (precision) or
diversity (recall). To mitigate this issue and
inspired by fixed-point optimization, a
penalty to the loss function of generative
models is introduced that minimizes
discrepancies between the model's weights when
trained on real versus synthetic data. Since
computing this penalty would require training
a new generative model at each iteration, a
permutation-invariant hypernetwork is proposed
to make evaluating the penalty tractable by
dynamically mapping data batches to model
weights. This ensures scalability and seamless
integration of the penalty term into existing
generative modeling paradigms, mitigating
biases associated with model autophagy.
Additionally, this penalty improves the
representation of minority classes in
imbalanced datasets, which is a key step
toward enhancing fairness in generative
models.
About the speaker:
Ali Siahkoohi is an incoming
tenure-track assistant professor in University
of Central Florida's Computer Science
Department. Currently, he is a Simons
Postdoctoral Fellow in the Department of
Computational Applied Mathematics &
Operations Research at Rice University, jointly
hosted by Dr. Maarten V. de Hoop and Dr. Richard
G. Baraniuk. He received his Ph.D. in
Computational Science and Engineering from
Georgia Institute of Technology in 2022. His
research focuses on designing scalable methods
for quantifying uncertainty in AI models, with a
broader goal of enhancing AI reliability.
June
25, 2025
A System
of Systems Cognitive Decision-Making
12:00 PM - 2:00 PM PT
Speaker:
Dr. Morantz
Register at: https://events.vtools.ieee.org/m/488111
Decision-making
is a task that an average person does about
300 to 400 times a day. Most decisions
are minor but there are some that are of great
importance, that the decision can have great
impact. The Butterfly Effect states that a
small action in one part of the world can
cause a great effect in another part of the
world at some later time. [Lorenz]
The Gartner Group estimates that by 2028 33%
of enterprise applications will include
agentic AI, and that this will enable 15% of
daily work decisions to be made autonomously,
without human intervention. [Gartner].
This can be fueled by a combination of
shortage of capable humans, an increase in the
cost of human involvement, and greater AI
accuracy and performance. It should be
started on a narrow realm of application, and
with knowledge, experience, and success, the
realm could be expanded. Human cognitive
function is an important part of this paper,
except that we try to create it in the machine
environment.
Some example situations are included to help
demonstrate the problem. This paper explains
some of the types of decision-making and how
they are performed. The paper then continues
with how this process, modeled after an
intelligent human would perform the task. This
discussion combines computer science, decision
sciences, psychology, and mathematics to
describe this project.
About the speaker:
Dr. Morantz, an IEEE Senior
Life Member, has a B.S. in C.I.S. and E.E., an
M.S. and Ph.D. in Decision Science, a mixture of
mathematical science including statistics,
psychology, and computer science. He has
additional course work in Computational
BioScience, Computer Science, statistical design
methodology, and Design Analysis Simulation
Experiments. Dr. Morantz has published and
presented on neural networks, multiprocessing
mathematics, biologically inspired computing
architecture including Artificial Intelligence
(AI), data-mining, and intelligent decision
making. His current research is in
biologically inspired computing for intelligent
decision making.
June 25, 2025
Applying Artificial Intelligence to Manage
Cost in Space Operations
8:00 AM PT
Speaker: Dr. Vince Socci
Register at: https://ieee-aess.org/presentation/webinar/applying-artificial-intelligence-manage-cost-space-operations
The Apollo Program was ultimately terminated due
to one fundamental issue: the high cost of
operations. In the new Space Age, the commercial
space industry's growth continues to be
constrained by cost. It is no surprise that the
primary launch companies rely on funding from
wealthy investors. For the space community to
expand and support a broader ecosystem of
participating companies, accurate cost estimation
and ongoing cost reduction are essential. Space
operations demand careful cost management
throughout the program lifecycle, from engineering
development to launch and recovery. Artificial
Intelligence offers new opportunities for managing
costs effectively. By leveraging data through
methods such as deduction, statistics, and machine
learning, we can achieve more accurate predictions
of production and operational costs. These
insights enable better cost decisions during
engineering development, production, and flight
operations. This lecture explores use cases for
applying AI to cost management in the space
industry and outlines cost management practices to
empower space economy entrepreneurs to reach for
the stars.
About the speaker:
Vince Socci is
the CTO of On Target Motion, where he provides
engineering, program management and business
development services for aerospace, automotive,
rail, marine, and other safety-critical
applications. Previously, as Product Cost Director
at Blue Origin, he managed the product
engineering, production, and operation cost of
rocket engines. Prior to that, he led National
Instruments transportation business development
throughout the Americas and provided business and
technical support for customers in vehicular
applications, with emphasis in propulsion and
autonomous systems. With 35 years of experience in
aerospace, automotive, rail, power electronics,
and medical systems, he has engineered systems in
the most complex applications. His specialized
areas of interest are embedded controls, real-time
test, and systems engineering for vehicle-based
applications. In the early 90’s, Socci designed
the first electronics for the Cummins B-series
diesel engine, which are still in use today. In
the mid-90’s, he developed power controllers for
GE locomotives. Late-90’s into 2000’s, he led the
development of the HybriDrive HEV powertrain,
which was used on various platforms from
commercial buses and taxis to military trucks.
Through the 2000’s into 2010’s, he led the
development of aero and auto vehicle control
systems for power, communications, fueling, radar,
motor controls, and unmanned systems. He was the
Director of Large Transport Fuel Systems for
Parker Aerospace, leading the development of the
A350XWB aircraft to first flight. Socci then
developed advanced validation systems for
propulsion and autonomous applications, using
simulation/emulation architectures, products, and
workflows to solve transportation product
development challenges. Currently, he is focused
on aerospace innovation including commercial space
transportation and UAV development. He is a Ph.D.
candidate and holds a BS in electrical
engineering, MS in electrical engineering and MBA
in technology management. Socci has served on the
Board of Directors and governing boards of several
professional societies, including IEEE, SAE, and
PMI. He also serves as an expert witness in
aerospace, automotive, and medical device
litigation.
August 19, 2025
The Sketches of Infinite Data
and Algorithms for Real-Time Data Insights
6:00 PM - 8:30 PM PT
Speaker: Dr. Vishnu S. Pendyala, San Jose
State University
Register at: https://events.vtools.ieee.org/m/482936
How are machine learning algorithms able to
answer questions from any nook and corner of the
World Wide Web? How are trending hashtags from
the near infinite microblog posts, unique
visitors and other distinct counts in the near
infinite website traffic determined? How do
blogging websites avoid recommending articles a
user has previously read? In general, how can we
answer complex queries about enormous data
streams without storing them entirely, in
real-time? The answer often lies in clever
approximation algorithms and data "sketches"
that capture essential properties using vastly
reduced space. The relentless flow of data in
modern systems indeed presents significant
challenges. These data streams are often too
large to store and too fast to process
exhaustively with traditional methods. This talk
introduces key sketching and approximation
techniques that help generate real-time data
insights by processing data streams.
About the Speaker
Vishnu S. Pendyala, PhD, is a faculty member in
Applied Data Science and an Academic Senator with
San Jose State University, current chair of the
Santa Clara Valley Chapters of IEEE Computer and
Computational Intelligence Societies, Area 4
Coordinator for Region 6, and a Distinguished
Contributor of the IEEE Computer Society. As a
past ACM Distinguished Speaker, researcher, and
industry expert, he gave nearly 100 talks and
tutorial sessions in various forums such as
faculty development programs, the 12th IEEE GHTC,
IEEE ANTS, 12th IACC, 10th ICMC, IUCEE, 12th ACM
IKDD CODS and 30th COMAD to audiences at venues
such as Stanford University, Google, University of
Bolton, Computer History Museum, Universidad de
Ingeniería y Tecnología, Lima, Peru, IIIT
Hyderabad, KREA, IIT Jodhpur, University of
Hyderabad, IIT Indore, IIIT Bhubaneswar. Some of
these talks are available on YouTube and IEEE.tv.
He is a senior member of the IEEE and ACM. He has
over two decades of experience in the software
industry in the Silicon Valley, USA. His book,
“Veracity of Big Data,” is available in several
libraries, including those of MIT, Stanford, CMU,
the US Congress and internationally. Two other
books on machine learning and software development
that he edited are also well-received and found
place in the US Library of Congress and other
reputed libraries. Dr. Pendyala taught a one-week
course sponsored by the Ministry of Human Resource
Development (MHRD), Government of India, under the
GIAN program in 2017 to Computer Science faculty
from all over the country and delivered the
keynote in a similar program sponsored by AICTE,
Government of India in 2022. Dr. Pendyala served
on a US government's National Science Foundation
(NSF) proposal review panel in 2023. He received
the Ramanujan memorial gold medal and a shield for
his college at the State Math Olympiad. He also
played an active role in the Computer Society of
India and was the Program Secretary for its annual
national convention.
|