Abstract: Autonomous vehicles are revolutionizing the current concept of transportation. Self-driving vehicles hold the potential to make travel safer, more efficient, and accessible to all. Their success relies on their ability to perceive and understand their environment, much like a human driver, and react upon the environment conditions to ensure that the vehicle is driven safely. However, despite the promise of sensing technology and its internal communication in the vehicle, it comes with its own set of limitations. Factors like adverse weather conditions, low-light situations, and distinguishing between objects of different sizes and materials can pose significant obstacles to achieve good and timely driving decisions. Moreover, there is the need to know more beyond the vehicle sensing boundaries to anticipate the vehicles’ decisions and self-control the different vehicles in the road. This is where Vehicle to Everything (V2X) communication steps in as a crucial complement, bridging the gaps left by these sensor limitations. V2X technology enables vehicles to interact with their environment, including communication with other vehicles and infrastructure elements, complementing their information with additional sensors. This type of communication facilitates the exchange of critical information, such as traffic conditions, road hazards, upcoming traffic lights and signs, and sensing information beyond the reach of the vehicle. Additionally, it promotes coordination between autonomous and human-driven vehicles, reducing conflicts and optimizing the traffic flow. This talk focuses on how vehicles communication between themselves, with infrastructure, people and 2-wheelers can improve and automate autonomous mobility. The current approaches employ edge based computing and cooperative sensing for fast object detection, V2X communication for disseminating detected objects, and seamless integration with vehicle control systems. This talk will also delve into real-world experiments with connected autonomous vehicles.

Biography:

Weblink:https://www.ua.pt/pt/p/10319259

Abstract: Today’s dominant AI paradigm—massive deep learning—treats perception as an offline classification problem, trained on static, exhaustively larger datasets, requiring multiple passes through the dataset. In contrast, the human visual system learns online, parsing a never-ending sensory stream into meaningful events, storing only what matters, and continually refining its internal model of the world. This lecture presents an alternate framework for AI that embraces these principles: self-supervised predictive learning on streaming video. Drawing on Event Segmentation Theory, cortical models of sequence memory, and our systems—STREAMER for hierarchical prediction and Predictive Attractor Models for future-state generation—I will show how a minimalist “predict-then-surprise” mechanism can (i) slice long, untrimmed videos into temporally coherent sub-events and (ii) spatially localize the actors driving those events, without a single manual label. Across canonical benchmarks (Breakfast Actions, 50 Salads, INRIA Instructional Videos) and a ten-day, 23-million-frame wildlife dataset, these models outperform prior unsupervised and weakly supervised methods and approach fully supervised accuracy, while learning in a single pass and storing no frames. By reframing perception as continuous prediction and rapid adaptation, we edge closer to the brain’s remarkable efficiency and open a path toward lifelong, label-free video understanding.

Biography: Sudeep Sarkar, Distinguished University Professor and Launch Dean Bellini College of AI, Cybersecurity, and Computing University of South Florida, Tampa

Weblink:https://cse.usf.edu/~sarkar/SudeepSarkar/About_Me.html

Abstract: Artificial Intelligence (AI) has rapidly evolved from a theoretical concept to a transformative force across industries and everyday life. Yet, for many, it remains an elusive and often misunderstood domain. This talk aims to demystify AI by tracing its journey from foundational ideas to real-world applications. We will explore core principles of AI, including machine learning, deep learning, and natural language processing, while addressing common misconceptions and ethical concerns. Through accessible explanations and practical examples—from healthcare and cybersecurity to personalized technologies and digital forensics—we will bridge the gap between complex algorithms and their tangible impact. Attendees will gain a clearer understanding of AI's capabilities, limitations, and the future possibilities that lie ahead, empowering them to engage more thoughtfully with this powerful technology.

Biography: Prof. Dr.S.S. Iyengar, Ph.D., D.Sc. (Hon.), Ph.D. (Hon.), Ph.D. (Hon.) ACM Fellow, IEEE Life Fellow, AAAS Fellow, NAI Fellow, AIMBE Fellow, SDPS Fellow, AAIA Fellow Member of the European Academy of Sciences, Member of the European Academy of Arts and Sciences Asian Digital Forensic Education and Research Director. Global Forensic and Justice Center Distinguished University Professor, Ryder Professor Director, US Army Funded Digital Forensics Center of Excellence

Weblink:https://people.cis.fiu.edu/iyengar/

Abstract: This is a general popular science talk on Generative Adversarial Networks, with a bit of history, two popular deep network architectures used in a GAN: from scratch, a very simple implementation using Perceptron-like single layer neural networks: first, hand-crafted, and then trained. The talk will have more physical significance than hard-core mathematics, and will have some take-home philosophies, more than mathematical details.

Biography:

Weblink:https://www.cse.iitd.ac.in/~sumantra/

Abstract: Adapting a model trained on vast amounts of data to new tasks with limited labeled data has long been a challenging problem, and over the years, a diverse range of techniques have been explored. Effective model adaptation requires achieving high accuracy through task-specific specialization without forgetting previous learnings, robustly handling the high variance from limited task-relevant supervision, and doing so efficiently with minimal compute and memory overheads. Recently, large language models (LLMs) have demonstrated remarkable ease of adaptation to new tasks with just a few examples provided in context, without any explicit training for such a capability. In this talk, we examine this emerging phenomenon and assess its potential to meet our longstanding model adaptation goals vis-a-vis methods like fine-tuning, adapters, and memory augmentation.

Biography:

Weblink:https://www.cse.iitb.ac.in/~sunita/

Abstract: Machine learning (ML) computations of increasing sophistication and complexity are being developed to solve complex, data-driven problems in di- verse areas. Their output is often subject to undesirable phenomena such as over- fitting and hallucinations that are hard to detect, and more generally their scientific rigor is hard to establish. We propose the concept of ML-solvability by combining the theories of learnability, computing and logic, which characterizes the model space, the learning algorithm that estimates a model using samples, and the inference algorithm that utilizes the model. It provides insights into the applicability and generalization of ML codes, and the possibility of incomplete and unsound inferences if the underlying problem is not ML-solvable. In science areas, the long-established laws are synergistically exploited to sharpen and com- pose powerful ML solutions with provable generalization and correctness properties. We describe a framework for ML-solvability and generalization analyses based on a combination of physical laws that govern systems and information laws that characterize the learning processes. We combine the learning dimension and training error to derive generalization equations to detect and minimize overfitting, and utilize physical law violations by learning processes to identify and mitigate inference inadequacies. We briefly describe the uses of smooth, non- smooth and algebraic forms of laws to develop or analyze ML solutions in the areas of data transport networks, nuclear engineering, and computer system diagnosis.

Biography: Nageswara (Nagi) Rao is a Corporate Fellow at Oak Ridge National Laboratory where he joined in 1993. He received B. Tech from National Institute of Technology, Warangal, India, M.E. from School of Automation, Indian Institute of Science, Bangalore, India, and PhD in computer science from Louisiana State University. His research areas include high performance and quantum networking, information fusion, machine learning, and federations of science instruments. He is a Fellow of IEEE, and received 2005 IEEE Computer Society Technical Achievement Award and 2014 R&D 100 Award.

Weblink:https://www.ornl.gov/staff-profile/nageswara-rao