Kyunghyun Cho on Statistics in the Era of Computation

On November 18, 2026, the Simons Foundation will host Statistics in the Era of Computation, a New York lecture by Kyunghyun Cho about learning the algorithms used to estimate statistical quantities. The program starts with standard deviation, a familiar measure of variation, before moving toward a broader argument about computation and statistical reasoning.

Kyunghyun Cho is a New York University professor and co-founder and CEO of Ortet. His lecture offers technically fluent researchers and AI operators a chance to examine a design choice that can disappear behind a model's reported performance: how the estimation method itself was built, and which problems it was built to handle.

The free, in-person Presidential Lecture runs from 6 to 7 PM ET at Gerald D. Fischbach Auditorium, 160 5th Avenue, New York. Advance registration is required, with a 2 PM cutoff on the day of the event. The relevant business question reaches beyond the lecture hall: what evidence should a team ask for when a learned method becomes part of its decision process?

Learning the estimator changes the design conversation

Kyunghyun Cho's announced program begins with training an estimation algorithm and then considers how the approach could extend to additional statistical quantities. Starting with standard deviation gives the audience a concrete example before the argument moves into computational identifiability, a concept the lecture will connect to a learned estimator and a prior distribution over problem instances. Those are the proposed subjects of the talk; the presentation and its supporting argument are still ahead.

An estimator turns observations into an estimate of something a researcher wants to understand. Once learning enters the design of that estimator, a new set of choices deserves attention: which example problems it sees, how its training objective is defined and where its performance is measured. Those choices can shape what a useful result means. An engineering team evaluating the method would need to examine them alongside the headline output.

That gives an audience member something specific to listen for. How is the distribution of problems specified? Which assumptions enter through that choice? How should a practitioner investigate a case that sits outside the training setup? These are editorial preparation questions prompted by the announced topic, rather than a promised session agenda. They let the discussion begin with a decision a technical team might actually face.

Compute makes statistical judgment more consequential

The wider research conversation already contains this tension. A 2025 Harvard Data Science Review perspective examines the relationship between statistics and data science, including differences in research culture and the importance of reproducibility, uncertainty and data quality. Its discussion places computational methods and statistical reasoning in the same working conversation. The paper supplies context for the lecture, without serving as evidence for an argument Kyunghyun Cho has yet to present there.

For an AI operator, the immediate implication is a better evaluation conversation. A favorable result is easier to use when a team understands the conditions that produced it. If the method was trained on a particular collection of problems, the evaluation should help explain how closely the intended application resembles those problems and where further testing is needed. Compute creates room to explore more elaborate methods; someone still has to decide which evidence supports their use.

The same responsibility appears at other points in software development. DevCuration's coverage of AI Coding Summit NYC's release decisions examines how agent-assisted changes earn approval. Here, the object under examination is an estimation method. The connection is the work of making a technical claim inspectable before it becomes an operating assumption inside a product or research program.

A researcher and company builder brings the question to New York

Kyunghyun Cho's current biography identifies him as the Glen de Vries Professor of Health Statistics and a professor of computer science and data science at NYU, alongside his role as Ortet's co-founder and CEO. NYU's faculty profile independently confirms his academic appointment and research interests in machine learning and artificial intelligence. That combination gives the lecture relevance to readers who work across research and company building.

His earlier company experience includes co-founding Prescient Design in 2021, followed by its acquisition by Genentech that year. According to his biography, his subsequent Genentech tenure ended in January 2026. Those details establish a history of working across academic and industrial settings. They offer a reason for technical founders to engage with his methodological argument, without turning the lecture into an announced Ortet product presentation or an endorsement by his former employer.

The Simons Foundation's Presidential Lectures are free public scientific talks. For this advanced subject, the closest audience fit is researchers, graduate students, ML engineers and leaders comfortable following statistical reasoning. A founder weighing attendance can bring an evaluation question from an actual project. The hour has a clearer purpose when that question is specific enough to survive contact with the mathematics.

Arrive with a question and the right registration

The organizer's Eventbrite registration sets out the entry requirements. Attendees must be 18 or older and register individually using the name shown on a valid photo ID. Bring that ID and the matching digital or printed Eventbrite ticket; walk-ins will be denied, and seating is first come, first served. The auditorium offers accessible seating.

Doors open at 5:30 PM, with no earlier entry, and admission closes at 6:20 PM. The venue is Gerald D. Fischbach Auditorium, 160 5th Avenue, New York, NY 10010. Check the organizer's current instructions and availability before traveling. The announced format is a lecture, so attendance planning should center on the scientific discussion rather than an assumed reception or investor program.

An engineering lead can make the preparation practical by choosing an estimation decision the team currently takes for granted. Write down the assumptions, the data used to assess the method and the case that would make the team hesitate. As Kyunghyun Cho moves from standard deviation toward learned estimation, that case gives the listener a place to follow the argument back into the next experiment.