Events

We study opinion dynamics on finite graphs. One of the classical examples is the voter model, in which each vertex is given some initial opinion, and opinions evolve with time by interactions between neighbouring vertices in which one side convinces the other. It is not hard to see that on a finite graph, the system will eventually reach a consensus, which can be seen as the output of this system.

In this talk we will introduce a new opinion dynamics model, which we call the Campaign model. In this model vertices interact as in the voter model but for a limited amount of time T ("the campaign"), and then a majority vote is taken ("elections"). The behaviour of this model depends on some geometric properties of the underlying graph.

We will discuss the behaviour of the Campaign model, and in particular ask at which times T the system loses correlation to the majority of the initial opinions. Next, we will study the dynamical noise sensitivity of the model, and ask when it exhibits a non-trivial threshold when noising the dynamics.

This is a joint work in progress with Gideon Amir, Omer Angel and Omri Marcus.

While neural networks revolutionized fields like computer vision and natural language processing, training them is time consuming and computationally demanding. Now, imagine that instead of training, one could simply search online for the most accurate model for their specific task and use it directly.  With more than 1.5 million public models available on HuggingFace, this is becoming feasible. However, retrieving the right model for each task is non-trivial, and the key technical challenge is designing meaningful representations of models themselves. To address this, we introduce ProbeLog, a probing-based approach for representing classification models from their weights. Then, by projecting CLIP text embeddings into this representation space we can enable zero-shot model retrieval.
 

Bio:
Jonathan is a Computer Science PhD student at The Hebrew University of Jerusalem, under the supervision of Prof. Yedid Hoshen.  Currently, He is working on weight space learning, aiming to understand what information can be extracted from pre-trained neural networks. His recent research can be viewed at https://pages.cs.huji.ac.il/jonkahana. 

The well-known Zig-Zag product and related graph operators, like derandomized squaring, are fundamentally combinatorial in nature. Classical bounds on their behavior often rely on a mix of combinatorics and linear algebra. However, these traditional bounds are not tight and frequently fail to align with experimental results. In this talk, we will present a more refined analysis that utilizes the full spectrum of the graph, rather than relying solely on its spectral expansion. This approach produces results that both match experimental observations and, in a sense, are proved to be optimal. Our technique is analytic, diverging from classical methods: for the upper bound, we apply finite free probability, while for the lower bound, we draw on results from analytic combinatorics.
Based on joint works with Itay Cohen, Gal Maor and Yuval Peled. No prior knowledge is required.

TBD

TBD

TBD

TBD