Invitation to Attend 

2023 IEEE Western New York Image and Signal Processing Workshop November 3rd, 2023 at the RIT Student Development Center

Registration Link

RIT Student Development Center Map Link

The Western New York Image and Signal Processing Workshop is a premiere venue for promoting research involving applications of AI, image and signal processing in the Western New York region and for facilitating interaction between academic researchers, industry professionals and students.

Featuring:
- Advanced techniques to uncover neural functions and anomalous patterns in the human brain,
- Advances in real-time imaging with two-photon microscopy,
- Advances in applications of A.I. technologies,
- A Panel Discussion on the Capabilities, Liabilities and Guardrails of LLMs & Generative A.I.

The Western New York Image and Signal Processing Workshop is presented in a full hybrid format for participants in the Western New York area; allowing in-person or remote (WebEx) presentation along with in-person or remote (WebEx) attendance. Poster presentations will be available as MP4 Video files during and for a short period after the Workshop.

Important Dates:

Workshop Cost:

Workshop Agenda:

Morning session: Submitted presentations, along with invited talks on Neuroscience advances in analysis, image & signal processing, statistical, and experimental design techniques to uncover neural functions of the human brain and development of techniques for identifying anomalous neural behavioral patterns (neuromarkers) based on the acquisition and analysis of inherently noisy multi-dimensional signals.

Afternoon session: Submitted presentations, along with an invited talk involving photonics, imaging and real-time computational processes, then invited talks on the capabilities, limitations, ethical risks, accountability, and guardrails of Large Language Models and Generative AI from four perspectives, followed by a panel discussion and debate on how these technologies may be optimized to augment, rather than detract from, the human experience.

Supporters & Sponsors:

  Center for Emerging & Innovative Sciences, University of Rochester

  Vanteon Wireless Solutions

  Rochester Section of the IEEE

  Rochester Chapter of the IEEE Signal Processing Society

  Rochester Chapter of the IEEE Computer Society

  Rochester Chapter of the IEEE Photonics Society

  Rochester Chapter of the Society for Imaging Science and Technology

Keynote and Invited Presentation Details:

Keynote presentation:

 Dr. John J. Foxe
 University of Rochester
 Research Director - The Ernest J. Del Monte Institute for Neuroscience
 Chair - Department of Neuroscience
 Kilian J. and Caroline F. Schmitt Chair in Neuroscience
 Professor - Department of Neuroscience, Dept. of Psychiatry, and Center for Visual Science


 Topic:
 Innovations in Neurophysiology and Approaches for the Development of Objective Inter-Species Disease Neuromarkers
Abstract:  Developing objective neuromarkers (endophenotypes) of disease in humans presents a unique challenge for researchers and clinicians in that, unlike most other organs in the body, we cannot readily take a biopsy of tissue or easily sample the fluids that bathe the brain. So, for example, it is only post-mortem that a definitive diagnosis of Alzheimer’s disease can be made when brain tissue can be examined for the telltale neurofibrillary tangles and amyloid-beta plaques. Nonetheless, developments in neuroimaging and neurophysiological approaches are allowing neuroscientists unprecedented ability to observe both functional and structural outcomes non-invasively, and much progress has been made in the development of objective measures for many diseases of the brain, both neurodevelopmental diseases and psychiatric conditions. In parallel, genetic engineering advances have allowed for the development of animal models of many of these human diseases, providing exquisite ability to measures disease processes in neural tissue. As extraordinary as this latter development has been, a second major challenge arises, and that is how to bridge the gap between the measures we can reasonably make non-invasively in human patients, with the detailed outcomes we are capable of in mouse models. In this presentation, I will describe some of the new technologies we have available in both the human and animal neurophysiological toolbox, and I will describe new approaches to bridging the inter-species “translational gap” through combined electrophysiological approaches.

Morning Invited presentations:

 Dr. Emily Knight, M.D., Ph.D.
 University of Rochester
 Assistant Professor - Department of Developmental & Behavioral Pediatrics, and Dept. of Neuroscience


 Topic:
 Uncovering Neural Signatures of Sensory Processing in Autism
Abstract:  Biological motion imparts rich information related to the movement, actions, intentions and affective state of others, which can provide foundational support for various aspects of social cognition and behavior. Given that atypical social communication and cognition are hallmark symptoms of autism spectrum disorder (ASD), many have theorized that a potential source of this deficit may lie in dysfunctional neural mechanisms of biological motion processing.  Synthesis of existing literature provides some support for biological motion processing deficits in autism spectrum disorder, although high study heterogeneity and inconsistent findings complicate interpretation. Here, we attempted to reconcile some of this residual controversy through the use of signal processing analysis.  This methodology allowed us to undercover a previously unrecognized modulating role for attention in biological motion processing in ASD. We employed high-density electroencephalographic (EEG) recordings while participants observed point-light displays of upright, inverted and scrambled biological motion under two task conditions to explore spatiotemporal dynamics of intentional and unintentional biological motion processing in children and adolescents with ASD (n=27), comparing them to a control cohort of neurotypical (NT) participants (n=35). Behaviorally, ASD participants were able to discriminate biological motion with similar accuracy to NT controls. However, electrophysiologic investigation revealed reduced automatic selective processing of upright biologic vs. scrambled motion stimuli in ASD relative to NT individuals, which was ameliorated when task demands required explicit attention to biological motion. Together, these data suggest that individuals with ASD are able to discriminate, with explicit attention, biological from non-biological motion but demonstrate diminished automatic neural specificity for biological motion processing. This may have cascading implications for the development of higher order social cognition.  The results highlight the superior sensitivity of neurophysiologic measurement over behavioral measures in identification of differential neural network recruitment in clinical populations.

  Dr. Ross Maddox
 University of Rochester
 Associate Professor - Department of Biomedical Engineering, and Dept. of Neuroscience


 Topic:
 Investigating attention across the auditory system using forward modeling of brain activity
Abstract:  Selective auditory attention—the ability to focus on one sound source in the presence of many others—is an essential part of humans’ ability to communicate verbally. Many studies have demonstrated a strong effect of attention on the way speech is encoded by the auditory cortex. The results of similar studies investigating attention in the auditory brainstem, which receives signals from the inner ear, extracts features such as pitch and location, and passes them on to the cortex, have been much more mixed. Part of what has made these studies difficult is that traditional methods for measuring brainstem responses involve thousands of repetitions of short, artificial sounds, which makes engaging attentional mechanisms difficult. Our lab has developed new methods for using natural, unrepeated speech to measure brainstem activity and investigate attention while listeners perform the ecologically relevant task of listening to one audiobook while ignoring another. Using these methods, along with custom electrodes which can be placed on the eardrum, we tested the effects of selective attention on the responses of distinct areas of the auditory system, from its initial encoding by the auditory nerve through the brainstem and cortex. We find a strong effect of attention in the cortex, replicating previous studies, but find no evidence of attentional effects in earlier areas of the brain.

  Dr. Samuel Norman-Haignere
 University of Rochester
 Assistant Professor - Department of Biostatistics nd Computational Biology, Dept. of Neuroscience, and Dept. of Biomedical Engineering


 Topic:
 Inferring the computations and organization of human auditory cortex
Abstract:  To understand the meaning of a sentence, recognize a familiar voice in a crowd, or pick out the melody in a song, the brain must rapidly recognize, remember, and synthesize information across timescales spanning milliseconds to seconds. Human auditory cortex is essential to this process, but many basic questions about its organization and computational properties have remain unanswered, in part due to the inherent challenge of probing responses to high-dimensional natural stimuli using coarse and noisy neuroimaging methods. In this talk, I will describe two methodological advances that make it possible to: (1) infer brain organization from responses to high-dimensional natural stimuli (2) test whether a computational model can explain a neural response by comparing natural and “model-matched” stimuli. By applying these methods to neuroimaging data and human intracranial recordings, we have uncovered some of the dominant response dimensions that organize human auditory cortex and inferred computational principles that underlie their response.

Afternoon Invited presentations:

  Dr. Michael Giacomelli
 University of Rochester
 Assistant Professor - Department of Biomedical Engineering, and Institute of Optics


 Topic:
 Real-time assembly of gigavoxel mosaics using two-photon microscopy during surgery
Abstract:  Two photon microscopy can image and diagnose cancer in fresh tissue, but is challenging to perform fast enough to provide real-time diagnosis during surgical procedures or biopsies where billions of voxels must be imaged in order to evaluate a specimen.  In this talk, I present an overview of the hardware and software solutions we developed to rapidly image very large (centimeter scale) tissues at sub-wavelength resolution and then assemble those images into a virtual histology that can be used to diagnose a patient or guide surgery in real-time.

  Jie An
 University of Rochester
 Department of Computer Science (Dr. Jiebo Liu, advisor)


 Topic:
 Generating Interleaved Image-text; a New Task in Generative AI
Abstract:   This work investigates a challenging task named open-domain interleaved image-text generation, which generates interleaved texts and images following an input query. We propose a new interleaved generation framework based on prompting large-language models (LLMs) and pre-trained text-to-image (T2I) models, namely OpenLEAF. In OpenLEAF, the LLM generates textual descriptions, coordinates T2I models, creates visual prompts for generating images, and incorporates global contexts into the T2I models. This global context improves the entity and style consistencies of images in the interleaved generation. For model assessment, we first propose to use large multi-modal models (LMMs) to evaluate the entity and style consistencies of open-domain interleaved image-text sequences. According to the LMM evaluation on our constructed evaluation set, the proposed interleaved generation framework can generate high-quality image-text content for various domains and applications, such as how-to question answering, storytelling, graphical story rewriting, and webpage/poster generation tasks. Moreover, we validate the effectiveness of the proposed LMM evaluation technique with human assessment. We hope our proposed framework, benchmark, and LMM evaluation could help establish the intriguing interleaved image-text generation task.

  Dr. Christopher Kanan
 University of Rochester
 Associate Professor - Department of Computer Science


 Topic:
 The Current Capabilities and Limitations of Foundation Models and Generative AI
Abstract:  Generative AI systems are now being used across all sectors of the economy. They have enabled fundamental advances in editing and drafting documents, programming, creating art, chat bots, automation, and much more. They have democratized skills that previously required years of training, and can greatly enhance the productivity of individuals and organizations. These systems have been dubbed “Foundation Models” because they are extremely large AI systems that can be used for many tasks, where these systems learn through self-supervised learning objectives. In this talk, I describe how some of the most influential systems work. I describe their current capabilities and limitations, and discuss some of the missing ingredients needed to further advance them.

  Dr. Jonathan Herington
 University of Rochester
 Assistant Professor - Department of Philosophy


 Topic:
 High-Level Ethical Risks for Generative AI
Abstract:  The rapid advancement of generative AI poses exciting possibilities but also raises critical ethical challenges around what it means to be a human being . There has been much focus on familiar ethical risks: (1) many models train on massive datasets scraped without consent, potentially undermining individual privacy and profiting off of public datasets, (2) training data reflects societal prejudices which can propagate unfair outputs that marginalize vulnerable groups, and (3) as systems become more autonomous, unpredictable behaviors beyond developer intentions could emerge. These risks are real and immediate, but generative models also raise more existential ethical questions: what does it mean to “create” something in the context of generative AI? Should we worry about the atrophy of creative and analytic skills in the face of tools that can do those tasks for most of us? What is it that humans are distinctively good at in a world with advanced generative models? In this talk, I explore a simple philosophical framework for thinking about these risks that focuses on (i) promoting human welfare, (ii) respecting human autonomy, and (iii) justly distributing the benefits and burdens of these systems.

 Dr. Andrew D. White
 University of Rochester
 Associate Professor - Department of Chemical Engineering


 Topic:
 Language is the future of Chemistry
Abstract:   Large language models (LLMs) are beginning a new era of chemistry, where LLMs can connect and reason about experiments, tools, and databases to accomplish open-ended tasks in chemistry. This is creating both new opportunities in accelerating chemistry, but also increasing the challenge of interpretation of model outputs. I will summarize our recent work on applying LLMs to authentic chemical challenges, their ability to use tools, access scientific literature, and how we can force them to explain themselves.

  Dr. Kathleen Fear
 University of Rochester
 Director of Data & Analytics, UR Health Lab
 Assistant Professor - School of Nursing


 Topic:
 Accountability and Generative AI in Healthcare
Abstract:   This presentation will explore the transformative potential of large language models (LLMs) and generative AI tools in addressing problems within the healthcare sector. The focus will be on leveraging these advanced technologies to streamline time-consuming procedures, reduce administrative burden, and enhance patient care. This talk will delve into how to effectively identify the appropriate problem spaces for the application of generative AI, with an emphasis on minimizing risks and ensuring the utmost safety and privacy for patients. Furthermore, we will touch on the ethical, regulatory, and accountability issues that come with the implementation of AI in healthcare.