MATRIX Fall Seminar Series – Dr. Vincenzo Lomonaco

On Friday, September 10, MATRIX is pleased to host Dr. Vincenzo Lomonaco as part of our seminar series. Fall seminars will be hybrid hosted via Zoom & in-person. Full details regarding the seminar may be found below –

Avalanche: An End-to-End Library for Continual Learning

Dr. Vincenzo Lomonaco

Artificial Intelligence and Continual Learning

University of Pisa

Friday, September 10, 2021

11 AM – 12 PM CST

https://utsa.zoom.us/j/98953233499

Abstract: Learning continually from non-stationary data streams is a long-standing goal and a challenging problem in machine learning. Recently, we have witnessed a renewed and fast-growing interest in continual learning, especially within the deep learning community. However, algorithmic solutions are often difficult to re-implement, evaluate and port across different settings, where even results on standard benchmarks are hard to reproduce. In this talk we present Avalanche, an open-source end-to-end library for continual learning R&D based on PyTorch. Avalanche is designed to provide a shared and collaborative codebase for fast prototyping, training, and reproducible evaluation of continual learning algorithms. Finally, we will discuss possible applications of such techniques in the real-world as well as future challenges in continual learning research that Avalanche may help us tackle.

MATRIX Fall Seminar Series – Dr. Ioannis Agadakos

Chameleons’ Oblivion: Complex-Valued Deep Neural Networks for Protocol-Agnostic RF Device Fingerprinting

Dr. Ioannis Agadakos

Northeastern University

Friday, September 17, 2021

11 AM – 12 PM CST

https://utsa.zoom.us/j/98953233499

Abstract: Prior work has demonstrated techniques for fingerprinting devices based on their network traffic or transmitted signals, which use software artifacts or characteristics of the underlying protocol. However, these approaches are not robust or applicable in many real-world scenarios. In this paper we explore the feasibility of device fingerprinting under challenging realistic settings, by identifying artifacts in the transmitted signals caused by devices’ unique hardware “imperfections”. We develop RF-DCN, a novel Deep Complex-valued Neural Network (DCN) that operates on raw RF signals and is completely agnostic of the underlying applications and protocols. We introduce two DCN variations: a retrofitted Convolutional DCN (CDCN) originally created for acoustic signals, and a novel Recurrent DCN (RDCN) for modeling time series. Our work demonstrates the feasibility of operating on raw I/Q data collected within a narrowband spectrum from open air captures across vastly different modulation schemes. In contrast to prior work, we do not utilize knowledge of the modulation scheme or protocol intricacies such as carrier frequencies. We conduct an extensive experimental evaluation on large and diverse datasets as part of a DARPA red team evaluation, and investigate the effects of different environmental factors as well as neural network architectures and hyperparameters on our system’s performance. Our novel RDCN consistently outperforms all baseline neural network architectures, is robust to noise, and can identify a target device even when numerous devices are concurrently transmitting within the band of interest under the same or different protocols. While our experiments demonstrate the applicability of our techniques under challenging conditions where other neural network architectures break down, we identify additional challenges in signal-based fingerprinting and provide guidelines for future explorations.

MATRIX Fall Seminar Series – Dr. Gabriel de Erausquin

Use of deep learning strategies to account for biological complexity and the genetics of neuropsychiatric disease

Dr. Gabriel de Erausquin

University of Texas Health San Antonio

Friday, September 24, 2021

11 AM – 12 PM CST

https://utsa.zoom.us/j/98953233499

Abstract:

Complex phenotypes present several challenges for genetic studies including the presence of epistasis, pleiotropy, and heterogeneity. We approached these problems in a data-driven fashion to test the hypothesis that the heritability expected from twin studies but unexplained by genetic studies is distributed in heterogeneous partitions of a complex trait, each with distinct genotypic-phenotypic associations. We designed a machine learning algorithm termed PGMRA to identify naturally occurring partitions in the data in an unsupervised fashion. PGMRA first dissects genome-wide data and uncovers a genotypic architecture composed of sets of variations shared by subsets of individuals. Next, phenotypic data are independently organized into natural sets of features such as clinical manifestations, voxels of neuroimages, or personality traits in a phenomic-like approach. Cross-matching of the two types of sets reveals multiple associations restricted to subgroups of individuals, thereby uncovering the genotypic-phenotypic architecture of a trait and accounting for its distributed genetic risk or propensity.

MATRIX Fall Seminar Series – Dr. Gabriela Ciocarlie

Everything is Connected: Security and Reliability for Critical Infrastructure

Dr. Gabriela Ciocarlie

University of Texas San Antonio

Friday, October 1, 2021

11 AM – 12 PM CST

https://utsa.zoom.us/j/98953233499

Abstract

This talk will present different approaches for security and reliability for critical infrastructure. Critical infrastructure often connects cyber and physical layers; in my work, I considered approaches that operate across both dimensions. The talk will focus on IoT, with automated methods to identify and monitor IoT devices, and cellular networks, with methods to detect anomalous cells and diagnose them.