Events

Human motion is a fundamental attribute, underlying human actions, gestures, and behavior. Research of human motion has a wide range of applications, from medical simulations, through security, to entertainment. While the domain of deep learning for RGB image analysis has received huge attention, the domain of motion manipulation using DNNs, and more recently diffusion models, is in its infancy, and holds many scientific opportunities yet to be discovered.
Moreover, motion is challenging due to its irregular structure, diverse angles, and the expense and complexity of obtaining high-quality data. My research focuses on generative tasks such as motion synthesis and motion editing using deep neural networks. In the upcoming talk, I will discuss the building blocks used for motion synthesis and elaborate on various works that utilize diffusion models for synthesis and editing tasks. Each work addresses different aspects of the motion domain and involves various controls and tasks.

Based on works of Busemann and Gromov, Anders Karlsson has proposed a theory of “metric functionals” on a general metric space, as a replacement for linear functionals when there is no linear structure. for discrete spaces, these identify with “horofunctions”. We will discuss some aspects of these objects for “well suited” metrics on discrete groups. We are mainly interested in the interaction between the geometry and the algebraic properties of the group. for example, it seems that metric functionals can be useful in studying the structure of “small” groups, and understanding the existence of (virtual) characters. I am not assuming familiarity with any of the above notions, and plan to properly define everything presented in the talk. This is based on joint works with Liran Ron-George.

Sparse solutions play a crucial role in various fields due to their compact representation, interpretability, robustness, and better generalization. This talk focuses on first-order methods for composite problems with an explicit sparsity expression, and the connection between the optimality conditions they achieve. I will show that the proximal-gradient operation can be implemented efficiently for classes of composite problems in spite of the sparsity element, and will present component-wise methods that achieve better solutions compared to the proximal-gradient. Finally, a conditional gradient based method that balances between the advantages and disadvantages of the proximal-gradient and the component-wise methods will be presented.

TBA

TBD

Lip-to-speech involves generating a natural-sounding speech synchronized with a soundless video of a person talking. Despite recent advances, current methods still cannot produce high-quality speech with high levels of intelligibility for challenging and realistic datasets. This talk presents LipVoicer, a novel method that generates high-quality speech, even for in-the-wild and rich datasets, by incorporating the text modality. Given a silent video, we first predict the spoken text using a pre-trained lip-reading network. We then condition a diffusion model on the video and use the extracted text through a classifier-guidance mechanism where a pre-trained automatic speech recognition (ASR) serves as the classifier. We demonstrate the effectiveness of LipVoicer through human evaluation, which shows that it produces more natural and synchronized speech signals than competing methods (demo page: https://lipvoicer.github.io). The presented LipVoicer is a joint work of Yochai Yemini, Aviv Shamsian, Lior Bracha, Sharon Gannot, and Ethan Fetaya.

Bio:

Sharon Gannot obtained his Ph.D. in electrical engineering from Tel-Aviv University, Israel, in 2000. He is a full professor in the Faculty of Engineering at Bar-Ilan University, Israel. He serves as a senior area chair for IEEE Transactions on Audio, Speech, and Language Processing, a member of the senior editorial board of IEEE Signal Processing Magazine, a member of the editorial board of IEEE SPS Education Center, and the chair of the IEEE Signal Processing Society Data Science Initiative. Previously, he was chair of the IEEE Audio and Acoustic Signal Processing Technical Committee in 2017–2018. He has also held other roles, such as the general co-chair of the 2010 International Workshop on Acoustic Signal Enhancement and the 2013 IEEE Workshop on Applications of Signal Processing to Audio and Acoustics. He is the general co-chair of Interspeech 2024, which will be held in September in Greece. His research interests include statistical signal processing and machine learning in the audio processing domain. The methods he develops utilize multi-microphone and multi-modal information. Applications include speech enhancement, noise reduction, speaker separation and diarization, dereverberation, speaker localization, and tracking. Sharon Gannot is the recipient of the 2022 European Association for Signal Processing Group Technical Achievement Award and a Fellow of the IEEE.

TBD

TBA

Typical time-efficient encoding and decoding algorithms for error correcting codes use linear space. We construct asymptotically good codes that can be deterministically encoded in almost linear time and sub-linear space, as well as asymptotically good codes that can be deterministically decoded in this complexity. The encodable codes are based on condenser graphs. The decodable codes are based on locally correctable codes and a new efficient derandomization method. We believe that the new derandomization method is of independent interest.

The talk is based on joint works with Joshua Cook (University of Texas at Austin).

A distribution is k-incompressible, Yao [FOCS ’82], if no efficient compression scheme compresses it to less than k bits. While being a natural measure, its relation to other computational analogs of entropy such as pseudoentropy (Hastad, Impagliazzo, Levin, and Luby [SICOMP 99]), and to other cryptographic hardness assumptions, was unclear.

We advance towards a better understating of this notion, showing that a k-incompressible distribution has (k-2) bits of next-block pseudoentropy, a refinement of pseudoentropy introduced by Haitner, Reingold, and Vadhan [SICOMP ’13]. We deduce that a samplable distribution X that is (H(X) + 2)-incompressible, implies the existence of one-way functions.

Joint work with Iftach Haitner and Jad Silbak.

Quantum computers are expected to revolutionize our ability to process information. The advancement from classical to quantum computing is a product of our advancement from classical to quantum physics -- the more our understanding of the universe grows, so does our ability to use it for computation. A natural question that arises is, what will physics allow in the future? Can more advanced theories of physics increase our computational power, beyond quantum computing?
An active field of research in physics studies theoretical phenomena outside the scope of explainable quantum mechanics, that form when attempting to combine Quantum Mechanics (QM) with General Relativity (GR) into a unified theory of Quantum Gravity (QG). QG is known to present the possibility of a quantum superposition of causal structure and event orderings. In the literature of quantum information theory, this translates to a superposition of unitary evolution orders.
In this talk we will show a first example of a computational model based on models of QG, that provides an exponential speedup over standard quantum computation (under standard hardness assumptions). We define a model and complexity measure for a quantum computer that has the ability to generate a superposition of unitary evolution orders, and show that such computer is able to solve in polynomial time two well-studied problems in computer science: The Graph Isomorphism Problem and the Gap Closest Vector Problem, with gap O( n^{1.5} ). 

The talk is based on https://arxiv.org/abs/2403.02937 .