From Data to Discovery: Hosted By Adaption Ambassadors
From Data to Discovery: The Model Adaptation Loop
A hands-on session for AI/ML engineers, researchers, developers, and builders working on model adaptation, data-centric AI, and improving the performance of their models through experimentation.
Training a model is only one part of the problem. In practice, the hardest questions often come before and after training: What data should the model learn from? How should that data change? What should we evaluate? And how do we use those results to decide what to try next?
From Data to Discovery explores the model adaptation loop as an iterative research process, where data generation, transformation, evaluation, and training continuously inform one another.
We’ll look at how Adaptive Data, Invent, and AutoScientist fit into this loop, what controlled experiments reveal about data and training recipes, and how Adaption can help automate parts of this process.
The session will move from the underlying problem to the research workflow, then into a live demonstration and hands-on exercise where you’ll apply the same thinking to a problem of your own.
This is not a session about simply fine-tuning a model. It is for people who want to understand how to systematically experiment with data, training, and evaluation to discover what actually improves a model.
What to Expect
A deep dive into why data is often the bottleneck in model adaptation and why changing the data can be as important as changing the training recipe
An exploration of Adaptive Data, Invent, and AutoScientist, and how they fit into an iterative model adaptation workflow
Findings from controlled experiments across data, recipes, and evaluation, including what we learned from measuring different approaches
A live Adaption demo, showing the workflow from defining a problem through experimentation and evaluation
A hands-on exercise where you’ll define a problem for your own model and think through how the adaptation loop could be applied
An open Q&A and discussion with time to bring your own models, datasets, experiments, and questions
Agenda
7:30 PM to 7:40 PM: Welcome and Introduction
7:40 PM to 8:05 PM: From Data to Discovery: Why Data Matters in Model Adaptation
[Carson Rodrigues], Adaption Ambassador
8:05 PM to 8:30 PM: The Model Adaptation Loop: Adaptive Data, Invent, and AutoScientist
[Carson Rodrigues] and [Prayag Dwivedi], Adaption Ambassadors
8:30 PM to 8:40 PM: What We Measured: Experiments Across Data, Recipes, and Evaluation
8:40 PM to 8:55 PM: Live Adaption Demo
8:55 PM to 9:00 PM: Q&A and Open Discussion
When & Where
Friday, October 2, 2026
7:30 PM to 9:00 PM IST
Online · Google Meet
The meeting link will be shared with registered attendees.
Who Should Join
This session is designed for:
AI/ML researchers and engineers
Developers working with foundation and adapted models
Builders experimenting with model training and evaluation
Researchers interested in data-centric AI and automated experimentation
Anyone working on a model adaptation problem and looking for a more systematic way to iterate
How To Register
Join the session to explore how the model adaptation loop can turn data, experimentation, and evaluation into a continuous discovery process.
Come with a model, dataset, experiment, or problem you’re working on. We’ll use the session to go beyond the theory, see the workflow in action, and explore how you can apply it to your own work.