לוח שנה משולב

Editing real images using a pre-trained text-to-image (T2I) diffusion/flow model often involves inverting the image into its corresponding noise map. However, inversion by itself is typically insufficient for obtaining satisfactory results, and therefore many methods additionally intervene in the sampling process. Such methods achieve improved results but are not seamlessly transferable between model architectures. Here, we introduce FlowEdit, a text-based editing method for pre-trained T2I flow models, which is inversion-free, optimization-free and model agnostic. Our method constructs an ODE that directly maps between the source and target distributions (corresponding to the source and target text prompts) and achieves a lower transport cost than the inversion approach. This leads to state-of-the-art results, as we illustrate with Stable Diffusion 3 and FLUX.

Bio:

Vladimir Kulikov, PhD student at the Technion, supervised by Prof. Tomer Michaeli. Currently studying Generative Models with emphasis on Computer Vision.

Abstract: Prioritizing computational principles offers a promising strategy to link neural activity, computation, and behavior. In this talk, I will focus on computational trade-offs as a core principle of brain function. Pareto optimality posits that systems optimized for multiple, competing goals are constrained to a low-dimensional manifold, the Pareto front, that captures the trade-offs shaping their organization. I will present theory and data supporting the Pareto framework: First, we apply this framework to large-scale whole-brain functional data in humans (resting-state fMRI dataset, N=1200) and demonstrate that individual differences in the brain's functional connectome lie on a robust triangle. We show that this triangle, interpreted using network analysis, clinical data, and task performance, reflects fundamental information-processing trade-offs. Second, we show that the Pareto front is an efficient representation for task-related brain dynamics. Third, we characterize the constraints of a control mechanism on Pareto manifolds, suggest a potential representation for it, and infer its possible breakdown points. Finally, we show evidence from ADHD and Alzheimer's disease supporting these theoretical predictions. If time allows, I will briefly present how ASD variation can be cast as a computational trade-off between accurate encoding and fast adaptation. Together, these findings demonstrate that trade-offs can account for diverse patterns of neural function and dysfunction, underscoring the Pareto framework's role as a key computational principle for understanding brain and cognition. 

AI systems are increasingly shaping people’s lives, opportunities, and scientific progress. But how can we trust the inferences of such complex, black-box systems? This question becomes even more urgent in the presence of two core challenges that are ubiquitous in high-stakes applications: data scarcity and test-time distribution shift. These issues not only limit the utility of AI systems but can also lead to misleading conclusions and unexpected failures.

In response to these challenges, this talk explores how fundamental statistical principles and modern ML can empower one another to enable trustworthy and practically useful inferences.

The first part focuses on reliable inference under limited data. I’ll introduce a framework that safely enhances the sample efficiency of any statistical inference procedure—such as conformal prediction and hypothesis testing—by adaptively leveraging synthetic data (e.g., from generative models). Crucially, this approach provides distribution-free error control guarantees without imposing any assumptions on the quality of the synthetic data. I'll demonstrate its broad applicability across diverse domains, from reliable protein structure prediction to principled win-rate evaluation of large reasoning models.

The second part enhances model robustness to drifting data. I'll introduce a new approach to test-time training, grounded in sequential statistical testing. Building on conformal betting martingales, I’ll first present a principled monitoring tool to detect data drifts. Using this tool, I’ll derive a rigorous ‘anti-drift correction’ mechanism grounded in (online) optimal transport principles. This mechanism forms the foundation of a self-training scheme that promotes invariance to dynamically changing environments. I'll outline the key ideas and expand on technical details, if time permits.

Bio

Yaniv Romano is an associate professor in the Departments of Electrical and Computer Engineering and Computer Science at the Technion. Previously, he was a postdoctoral scholar in the Department of Statistics at Stanford University. Yaniv holds a PhD, MSc, and BSc in Electrical Engineering, all from the Technion. His super-resolution technology, invented with Peyman Milanfar, has been integrated into Google’s flagship products, including the Pixel phone. His uncertainty quantification technique, developed with Emmanuel Candes, was employed by The Washington Post to estimate outstanding votes during the U.S. presidential election.

Yaniv has received several honors and awards, including the ERC Starting Grant, the SIAG/IS Early Career Prize, the Sheila Samson Prime Minister’s Prize (Researcher Recruitment Prize), the IEEE Signal Processing Society Best Paper Award, the Alon Scholarship, the Krill Prize for Excellence in Scientific Research, and the Henry Taub Prize for Academic Excellence. Yaniv is a member of the Young Israel Academy.

The Duflo Serganova (DS) functor is probably the most fun (okay, also useful) tool that we have in super representation theory.   I will explain some work which attempts a meagre generalisation of this functor to modular representations of algebraic groups, which we call One Tree Island (OTI) functors.  For this we will need the exotic tensor category Ver_p which arises as the semisimplification of Rep(Z/p).  OTI functors share some properties of DS functors but also seem to be more complicated.  I will discuss two interesting cases of symmetric groups and reductive algebraic groups, and how in these cases OTI reproduces known functors of interest in a simpler way.