Symposium 2025

Michael shares the story of how he launched two high-growth startups—first LULA, then Gail—starting from his college dorm room. Through candid reflections on risk-taking, building in regulated industries, and lessons from the journey, he'll inspire you to stop waiting for permission and start building.

Tony Barr will discuss the early history of the SAS system for data analytics and his current work, A Model Of Reality (AMOR). AMOR aims to create a world where young children and adults can understand and flourish in programming, databases, mathematics, and data analytics. AMOR uses diagrams and flowcharts to make reading and writing programs easy and intuitive for learners and seasoned users. The mission is to enable users to navigate through knowledge space naturally, as they navigate the real world.

Ever since November 2022 when ChatGPT was released commercially, businesses have been pursuing the best approaches to implement AI into their processes. There has been a great deal of Fear of Missing Out (FOMO) related to this quest for value out of AI capabilities. In this talk, we will discuss what is working and what is not. And, we will explore how students can best position themselves for roles in AI as the technology continues to evolve in business.

Drawing from over 20 years in AI and NLP, Sarah explores emerging trends in generative AI, the challenges of building inclusive datasets, and the importance of supporting low-resource languages. With insights from both industry and research—including her work on IBM Watson's Jeopardy! Challenge—she highlights how responsible AI depends on diverse data, ethical practices, and a balance between scientific rigor and creativity.

Assumptions are a necessary part of making conclusions and inferences from data. Formal training in Statistics and data science tends to encourage a mathematical and automatic treatment of assumptions, with relatively low expectations for justifying assumptions in the context of the problem or conveying the extent to which conclusions are conditional on assumptions. With so much focus on methods and computing, it is important to also take time to step back and more deeply consider the layers of assumptions that make up the foundation of any data analysis. I believe statisticians and data scientists have a responsibility to better convey the conditional nature of results, and to work with subject matter experts to translate and interrogate assumptions within a particular scientific context. While this is challenging in most problems, it is a necessary step toward better justifying the use of, and trust in, statistical results. The goal of this high-level talk is to increase, or re-highlight, awareness and spur discussion about related challenges and strategies.

This talk introduces my recent e-book with the same title, freely available on arxiv. The book uses the modern theory of artificial intelligence (AI) to understand human suffering or mental pain. Both humans and sophisticated AI agents process information about the world in order to achieve goals and obtain rewards, which is why AI can be used as a model of the human brain and mind. The book starts with the assumption that suffering is mainly caused by frustration. Frustration means the failure of an agent (whether AI or human) to achieve a goal or a reward it wanted or expected. Frustration is inevitable because of the overwhelming complexity of the world, limited computational resources, and scarcity of good data. In particular, such limitations imply that an agent acting in the real world must cope with uncontrollability, unpredictability, and uncertainty, which all lead to frustration. Such computational theory is finally used to derive various interventions or training methods that will reduce suffering in humans. The ensuing interventions are very similar to those proposed by Buddhist and Stoic philosophy, and include mindfulness meditation.

While deep neural networks are powerful tools, they are known to have security issues. One particular threat is the so-called Trojan attack, which can be used to compromise, for instance, self-driving cars. I will describe our approach for detecting such attacks on vision models presented at NeurIPS' 21. This approach is motivated by basic neuroscientific principles and uses methods of topological data analysis (TDA), which I will introduce along the way. (No previous background in topology is necessary.)

The Bayesian approach to data analysis provides a powerful way to handle uncertainty in all observations, model parameters, and model structure using probability theory. Probabilistic programming languages make it easier to specify and fit Bayesian models, but this still leaves us with many options regarding constructing, evaluating, and using these models, along with many remaining challenges in computation. Using Bayesian inference to solve real-world problems requires not only statistical skills, subject matter knowledge, and programming, but also awareness of the decisions made in the process of data analysis. All of these aspects can be understood as part of a tangled workflow of applied Bayesian statistics. Beyond inference, the workflow also includes iterative model building, model checking, validation and troubleshooting of computational problems, model understanding, and model comparison. We review all these aspects of workflow in the context of several examples, keeping in mind that in practice we will be fitting many models for any given problem, even if only a subset of them will ultimately be relevant for our conclusions.

As artificial intelligence (AI) becomes increasingly integrated into products and services across industries, ensuring its responsible development and deployment has become a critical challenge. Several tools, processes, and principles have been developed to support responsible AI (RAI) in industry practice. However, research has shown a persistent gap between the aspirational goals of these RAI interventions and the practical realities faced by practitioners tasked with designing and building responsible AI systems. In this talk, I will first present a set of insights gained from empirical studies with industry RAI practitioners. I will then share a series of tools and processes designed to better support RAI practices in real-world industry settings—particularly in the contexts of AI auditing, red-teaming, and impact assessment. Finally, I will briefly discuss future directions for research, practice, and policy in building safe and responsible AI within industry.

This talk explores how we can make AI systems faster and easier to understand by bridging explainability and performance. It introduces a new framework that integrates logic-based neural models with custom hardware acceleration to enable single-cycle inference. Through this approach, the talk demonstrates how we can build AI systems that are not only highly interpretable but also capable of operating in real-time environments.

With over two decades at Microsoft and Google, Carlos offers an inside look at building a career across the world's top tech companies. He'll share lessons on leadership, navigating industry shifts, and what it truly means to drive innovation and transformation at scale.

In this reflective talk, Luciane shares key lessons from her three-decade journey at Microsoft—from career milestones to daily life in tech. She'll touch on the evolution of the IT industry, her core values, and practical advice for navigating a successful and fulfilling career in tech.

In this talk, Tyler shares lessons learned from his journey breaking into data science—from early rejections to landing roles at Facebook, Streamlit, and Snowflake. He'll offer practical tips for staying resilient, building projects that stand out, and navigating an unpredictable career path with curiosity and creativity.

This talk will go over an AI crash course, AI research in practice, and tools and resources for getting started with AI research.

This introduction workshop will provide you a foundational understanding of State Space Models (SSMs) and their applications in time-series modeling.

What happens when what we know isn't enough to be employable anymore? Amid rising layoffs in the tech industry and beyond due to the automation of labor with AI systems, stress about job security is thriving -- but hope for a meaningful, stable career is not yet lost. Attend this workshop to explore how analytical, generative, and agentic AI systems are being implemented in the modern workplace and understand their strengths and limitations. Learn about not just philosophical but technical arguments describing what exactly an AI system can and cannot provably do -- and why what makes us human is not only 'marketable' but essential to the current and future workforce.

Signals are how computers (and people) interpret the world. With the LLM hype calming down, we get a chance to revisit the mathematical roots of black-box machine learning in time-series problems, and learn how filters gave humanity access to computational foresight. Attend this workshop to attain mathematical insight into the complex world of signal processing from the perspective of its innovators, and learn what the future of AI beholds.

This workshop introduces fundamental Natural Language Processing concepts through the lens of medical text embeddings. Participants will learn how to transform unstructured clinical text into meaningful vector representations. We cover text preprocessing techniques, tokenization, normalization, and various embedding methods from statistical approaches to neural representations.

Learn how to properly configure a CV model using sophisticated annotation platforms and a live demo of YOLO training. Hands-on participation where the audience can help train a gator detector model.