IdoAI Events & Archive

Abstract: Responsible investing aims to generate positive impact across Environment (E), Society (S), and Governance (G), and rating companies along these dimensions is now widespread, making ESG scores highly popular. Current ESG solutions cannot model high-dimensional, multimodal time series capturing the joint evolution of ESG risks, financial returns, news, and sentiment, even though this domain requires jointly reasoning over distinct numerical signals where both numerical proximity and semantic type must be preserved. In this talk, we discuss a novel domain-aware representation learning framework that learns geometry-preserving representations for heterogeneous time series. The resulting model jointly learns to forecast future values and to organize entities in a representation space that reflects their temporal evolution. Trained on ESG, returns, news, and sentiment, the domain-aware LLM learns a representation space that enables downstream applications such as latent-space search for superior asset selection and portfolio rebalancing.

Abstract: Understanding human behaviour is essential for effective communication and social interaction. Gestural cues such as eye gaze, facial expressions, and body movements offer deep insights into complex mental and emotional states. Accurately capturing and modelling these cues in unconstrained environments remains a significant challenge due to data annotation complexity, variability in human behaviour, and computational limitations. My research focuses on developing robust computational models that can automatically learn and interpret non-verbal gestural cues in naturalistic, real-world scenarios, with minimal supervision. These models aim to bridge the gap between human cognitive understanding and machine perception, making AI systems more socially aware and context-sensitive.

My work has wide-ranging applications in real-world settings, including monitoring student engagement in classrooms, assessing group cohesion in collaborative environments, detecting emergent leaders, identifying dominant individuals in discussions, and analysing crowd behaviour. By advancing automatic behaviour understanding 'in-the-wild', my research contributes to building more adaptive and human-centric AI systems capable of functioning effectively in everyday social environments.

Abstract: Obscuring sensitive data from adversarial eavesdropping in distributed decision-making is a significant challenge across a range of applications, including smart grids, autonomous vehicle coordination, and multi-robot surveillance. Existing methods—such as encryption or differential privacy—modify transmitted messages without considering the underlying dynamics of the cyber-physical processes that generate the data, nor the capabilities of potential adversaries. In our recent work, we demonstrate that every dynamical system possesses an inherent system-level privacy that can be harnessed through the design of intelligent inter-agent communication protocols.

In this talk, I will present our recent results on extracting system-level privacy in the presence of eavesdropping adversaries. We derive analytical expressions that quantify the level of protection as a function of both the adversary's capabilities and the agent communication graph. Finally, I will show how these results can be applied in the context of distributed optimization and federated learning.

Speaker Bio: Dipankar Maity is an Assistant Professor in the Department of Electrical and Computer Engineering at the University of North Carolina at Charlotte. He earned his B.E. degree in Electronics and Telecommunication Engineering from Jadavpur University, India, in 2013, and his Ph.D. degree in Electrical Engineering from the University of Maryland at College Park, in 2018. Following his doctoral studies, he was a Postdoctoral Fellow at the Georgia Institute of Technology. He has been a visiting scholar at the Technical University of Munich, Germany, and at the KTH Royal Institute of Technology, Sweden. He is a member of the IEEE Control System Society Technical Committee on Security and Privacy. He is an Associate Editor for the IEEE Control Systems Letters journal.

Abstract: In December 2023, a Chevrolet dealership chatbot agreed to sell a car for $1, even declaring it a “legally binding offer.” While the dealership quickly corrected the error, the system was doing exactly what it was designed to do: optimise a loss function over token sequences, without any concept of margins, compliance, or risk.

This incident underscores a critical challenge for enterprises: the disconnect between probabilistic AI models and deterministic business imperatives. Generative AI does not “think” in terms of revenue, liability, or brand, it optimizes probabilities in a mathematical space and subjects them to the laws of chance.

This talk demystifies the hype and exposes the engineering reality. We will explore how Large Language Models rely on high-dimensional geometric transformations rather than logical reasoning. When a model “hallucinates,” it is not malfunctioning, it is extrapolating into sparse regions of its training manifold where patterns are unreliable. This is a structural property of the model, not a coding flaw.

We will connect these principles to real-world deployment strategies through three business-focused case studies:

• The Alignment Problem: Why semantic similarity fails to enforce business logic and compliance.
• The Architecture of Failure: Two examples where economic incentives collide with optimisation dynamics, resulting in costly missteps.
• A New Deployment Framework: Practical methods to align stochastic model behaviour with deterministic business goals.

Our goal is to equip decision makers with a mental model grounded in both mathematics and market reality. The organisations winning with GenAI are not those treating it as magic, they are those engineering it to serve strategy, safeguard margins, and scale responsibly.

Abstract: Recent crises, such as wildfires in the western USA, underscore complex and non-obvious interrelationships between human actions and impacts. Wildfire poses an increasingly serious threat due to a combination of climate change and human factors. During a wildfire, evacuation is commonly recommended to reduce the risk of injury or death. Although ideally an evacuation should finish as quickly as possible, certain traffic patterns that occur during an evacuation can cause significant delays. A better understanding of evacuation processes is needed to determine which actions may alleviate the congestion that causes these delays.

Agent-based models (ABMs) are powerful tools for modeling complex human dynamics at an individual level. Spatial agent-based models can be scaled up to handle large study areas, massive agent numbers, and complex agent-environment interactions needed to address real world challenges. Due to a lack of wildfire evacuation data, many evacuation models rely heavily on heuristics about evacuee behavior, especially for inferring evacuee wayfinding and routing decisions. Extensive investigations assume that an evacuee takes the shortest or quickest path to the nearest exit. However, this assumption poorly matches both surveyed evacuee preferences and congestion patterns in real-world evacuation traffic.

Using an agent-based model of vehicle evacuation, this study assesses the effects of choosing different routing strategies on evacuation clearance times and traffic patterns. Specifically, driving strategies that preferentially select the shortest path, the quickest path, and a path more likely to contain major roads are compared. The results indicate that there are marked differences between clearance times and evacuation curves, as well as different spatiotemporal patterns of traffic congestion between these strategies. These findings suggest that the shortest or quickest path routing strategy may underestimate evacuation durations and cause areas of high congestion in less likely locations and times.

Speaker Bio: Dr. Rebecca C. (Becky) Vandewalle is an incoming Assistant Professor in the School of Information at the University of South Florida, starting in Fall 2025. She graduated from the University of Illinois Urbana-Champaign in 2025 with a PhD degree in Geography. Dr. Vandewalle earned her MSc in GIS and Archaeology from the University of Edinburgh, completed a Post-Baccalaureate program in Classics at the University of Pennsylvania, and earned her Bachelors degree in Greek and Roman Studies from Rhodes College. Her research interests include crisis informatics, data science and decision support in complex socio-technical contexts, high-performance computing, and large-scale geospatial simulation.

Abstract: Due to their extraordinary ability to coordinate large numbers of degrees of freedom, soft-bodied marine animals, especially octopuses, remain great sources of amazement, mystery, and inspiration. Their softness challenges our understanding of controlling highly flexible systems, while their distributed neuroanatomy raises questions about sensory information processing and motor control. Insights from octopus studies can inspire advancements in soft robotics and medical technology, offering transformative benefits across multiple fields.

In this talk, I will present our work on an octopus-inspired soft arm, including modeling, estimation, and various control strategies (open-loop and feedback). First, I will present a mathematical model of a soft muscular arm using the formalism of Cosserat rod theory. The flexible rod dynamical equations are understood in terms of the stored energy content (potential energy), leading to an energy-shaping control (open-loop) method that accomplishes various manipulation tasks. Next, I will show how energy-based ideas can be used in shape estimation of slender soft-bodied systems from video recordings, with applications in soft robotics and octopus arms. Finally, I will introduce the fascinating ways octopuses are believed to use their various sensory capabilities in target-oriented motor control. I will finish by presenting a novel bioinspired sensory feedback control strategy that may explain stereotypical octopus arm movements.

Speaker Bio: Dr. Udit Halder is an Assistant Professor in the Department of Mechanical and Aerospace Engineering of the University of South Florida. Prior to joining USF, he was a postdoctoral research associate at the University of Illinois Urbana-Champaign. Dr. Halder earned his Ph.D. and M.S. degrees in Electrical Engineering from the University of Maryland, College Park; and his B.E. degree in Electronics and Telecommunication Engineering from Jadavpur University, Kolkata. His research interests range from the collective behavior of natural flocks to the motion of soft-bodied marine animals, with applications in robotics. His research has been published in prestigious journals and conferences, and it has received coverage from a variety of media outlets.

Abstract: Biodiversity crisis, steering the direction towards the sixth mass extinction. The rate at which species are disappearing now outpaces the rate at which we're discovering them. Insects, the most diverse group on Earth, are particularly under threat, with beetles boasting the highest number of known species. However, a shortage of trained taxonomists means many lesser-known insect groups remain largely unexplored.

Artificial intelligence (AI) offers a promising path forward for biodiversity exploration. While current apps like iNaturalist use image-based AI models to successfully identify well-known insect groups, they struggle with species for which data is scarce. Specialized AI models could overcome this by focusing on specific morphological features like antennae, wing venation, elytra, and reproductive structures. This would be particularly beneficial for ecologically vital insect groups such as beetles, bugs, and bees. Therefore, developing such a specialized AI model is crucial right now.

Insects are undeniably our allies. They play critical roles in pollination, nutrient cycling, and broader ecological processes. Pollination, far from just a fascinating natural phenomenon, is an essential function for survival. Without pollinators, neither humans nor Earth's terrestrial ecosystems could persist. Out of the crop plants grown globally, which provide all our food and plant-based industrial products are nearly 80% depend on animal pollination. Bee visits and other pollinator activities also lead to larger, more flavorful fruits and significantly higher crop yields. To better understand these vital processes, quantum dots are emerging as a revolutionary tool. These tiny particles glow brightly under UV light, allowing us to label and trace individual pollen grains. However, the sheer volume of data generated i.e, images of pollen on stigmas, on insect bodies, and in the air, is too vast for manual analysis. This is where AI, particularly computer vision and machine learning, becomes indispensable. AI can analyze this deluge of data, helping us to unravel the complex dynamics of pollination.

Conversely, insects can also be our adversaries, acting as noxious pests that inflict immense agricultural and economic losses. To ensure a sustainable future, it's vital to forecast their spatial and temporal distribution patterns. Species distribution models can predict current and future insect distributions using climatic and environmental data. Similarly, Agent-based models are modern tools for simulating complex scenarios like pest-host interactions. These models rely on predefined parameters such as birth rates, death rates, feeding rates, and migration rates within a spatially and temporally explicit context.

Speaker Bio: Dr. Debika Bhunia is an entomologist with a Ph.D. from the University of Kalyani and the Zoological Survey of India, focusing on the taxonomy and identification of phytophagous scarabs. She has been awarded the CSIR-SRF Direct 2024 and the "Ernst Mayr Travel Grants in Animal Systematics" from Harvard University's Museum of Comparative Zoology, Student Grant from Punjab University, best oral presentation at Animal Taxonomic Summit 2023, By MOEFCC. An active member of the Royal Entomological Society, London. Dr. Bhunia boasts an impressive publication record of 40 papers, two books, an h-index of 6, and 127 citations, and is credited with the discovery of 20 new species of Indian phytophagous beetles. She has also conducted numerous field surveys throughout various districts of West Bengal, India. She has completed research visits to prestigious institutions such as Museum Alexander Koenig in Bonn, Germany and Muséum National d'Histoire naturelle in Paris, France. Her discoveries received significant media attention in leading newspapers like Statesman, Times of India, Bartaman, Ananda Bazar Patrika, Press Trust of India and Press Information Bureau (PIB).

Background & Details: A group meeting with Mr. Mohammad Saif in the AI Board Room on applications of artificial intelligence to critical challenges in the Indian railway sector.

Speaker Bio: Mohammad Saif is Chief Mechanical Engineer for Planning and Projects at North Central Railways of Indian Railways. He graduated in Mechanical Engineering from IIT-Roorkee (formerly University of Roorkee) and holds a Masters Degree in Railway Engineering from KTH Stockholm. He is credited with the design of special freight vehicles for the Ministry of Defence, specialized wagons for the steel industry, and the indigenous development of a quick draw gear for slackless draw bars. He has also served as a member of the National Security Council Secretariat under the Ministry of Defence, where he contributed to strategic railway research. His research interest lies in railway vehicle dynamics, development of high speed bogies for freight vehicles and the application of artificial intelligence to address critical challenges in the Indian railway sector.

Abstract: In this talk, I will articulate a vision of artificial intelligence as “normal technology,” in contrast to both utopian and dystopian visions that treat AI as a potentially superintelligent entity. I will argue that the impacts of advanced AI, even if transformative, will unfold slowly, making a critical distinction between AI methods, AI applications, and AI adoption. I will discuss a potential division of labor between people and AI in a world with advanced AI and examine the implications of AI as normal technology for AI policy, AI safety, and human progress.

Speaker Bio: Sayash Kapoor is a Senior Fellow at Mozilla, a Laurance S. Rockefeller Fellow in the Princeton University Center for Human Values, and a computer science Ph.D. candidate at Princeton University's Center for Information Technology Policy. He is a coauthor of AI Snake Oil, one of Nature's 10 best books of 2024. His newsletter by the same name is read by over 50,000 AI enthusiasts, researchers, policymakers, and journalists. His work has been published in leading scientific journals such as Science and PNAS. He has written for mainstream outlets including The Wall Street Journal and WIRED and his work has been featured in The New York Times, The Atlantic, Washington Post, Bloomberg, and many others. Kapoor has been recognized with various awards, including a best paper award at ACM FAccT, an impact recognition award at ACM CSCW, and inclusion in TIME's inaugural list of the 100 most influential people in AI.