שני הסמינרים נחשבים כשמיעת סמינר אחד - צריך לחתום נוכחות בשניהם כדי לקבל קרדיט שמיעה
הרישום לסמינר יבוצע באמצעות סריקת הברקוד למודל (יש להיכנס לפני כן למודל, לא באמצעות האפליקציה)- הרישום מתחיל ב- 14:55 ומסתיים ב- 15:10
Registration to the seminar will be done by scanning the barcode for the Moodle (Please enter ahead to the Moodle, NOT by application)- Registration ends at 15:10
Electrical Engineering Systems Seminar
Speaker: Sarah Shitrit
M.Sc. student under the supervision of Dr. Ilai Bistritz
Wednesday, 4th June 2025, at 15:00
Room 011, Kitot Building, Faculty of Engineering
Cordial Learning: Distributed Training with Correlated Data
Abstract
We consider a distributed learning task with agents that have correlated labels. Specifically, the label of an agent depends on the input of other agents for the same sample. Correlated labels are the reality when agents share the same environment. Existing decentralized methods, such as federated learning, ignore the correlated data and perform poorly. On the other hand, centralized approaches are infeasible due to privacy and communication constraints. We introduce cordial (correlated and distributed) learning to address this gap by sharing only low-dimensional outputs between the agents while training local modules to extract informative signals from peers. This distributed learning induces a game, where each agent’s loss depends on the models of others. Assuming a linear model, we prove that cordial learning converges with probability one to a globally optimal solution, despite the global objective being nonconvex. Experiments on structured multi-digit MNIST tasks demonstrate that Cordial Learning remains robust even in highly nonlinear settings.
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
הרישום לסמינר יבוצע באמצעות סריקת הברקוד למודל (יש להיכנס לפני כן למודל, לא באמצעות האפליקציה)- הרישום מתחיל ב- 15:30 ומסתיים ב- 15:40
Registration to the seminar will be done by scanning the barcode for the Moodle (Please enter ahead to the Moodle, NOT by application)- Registration ends at 15:10
Electrical Engineering Systems Seminar
Speaker: Ariel Kantorovich
M.Sc. student under the supervision of Dr. Ilai Bistritz
Wednesday, 4th June 2025, at 15:30
Room 011, Kitot Building, Faculty of Engineering
Up Your Game: Training Games with Efficient Nash Equilibria with Deep Learning
We consider a game with N cooperative players that have a global objective.
Simple distributed algorithms such as gradient play often converge to a Nash equilibrium (NE).
However, a NE typically suffers from poor global performance. Converging to the social optimum requires explicit communication and coordination.
In this paper, we propose a new approach to improve the performance at NE. Our method uses machine learning offline training to design games with efficient NE. In particular, we use a dataset of games with parameters coming from a certain distribution (e.g., uniformly random player locations).
We then train a deep neural network (DNN) where the input is the local measurement available to the player and the output is the reward parameters that best approximate the global objective. We demonstrate our approach for three classes of games: quadratic games, energy games, and
wireless power control games. Our approach offers significant performance boosts while requiring no communication between the players and no complexity increase in real-time.
