October 26, 2021
Pavilion 10, EXPO Tel Aviv
Image Denoising – Not What You Think
Laurence Moroney leads AI Advocacy at Google. He's the author of over 20 books, including the recent best-seller "AI and Machine Learning for Coders" at O'Reilly. He's the instructor of the popular online TensorFlow specializations at Coursera and deeplearning.ai, as well as the TinyML specialization on edX with Harvard University. When not googling, he's also the author of the popular 'Legend of the Locust' sci fi book series, the prequel comic books to the movie 'Equilibrium', and an imdb-listed screenwriter. Laurence is based in Washington State, in the USA, where he drinks way too much coffee.
'State of the Union' Around the TensorFlow Ecosystem
Yoav Shoham is professor emeritus of computer science at Stanford University. A leading AI expert, Prof. Shoham is Fellow of AAAI, ACM and the Game Theory Society. His online Game Theory course has been watched by close to a million people. Prof. Shoham has founded several AI companies, including TradingDynamics (acquired by Ariba), Katango and Timeful (both acquired by Google), and AI21 Labs. Prof. Shoham also chairs the AI Index initiative (www.AIindex.org), which tracks global AI activity and progress, and WeCode (www.wecode.org.il), a nonprofit initiative to train high-quality programmers from disadvantaged populations.
Lessons from Developing Very Large Language Models
Mark Grobman is the ML CTO at Hailo, a startup offering a uniquely designed microprocessor for accelerating embedded AI applications on edge devices. Mark has been at Hailo since it was founded in 2017 and has overseen the ML R&D in the company. Before joining Hailo, Mark served at the Israeli Intelligence Corps Technology Unit in various inter-disciplinary R&D roles. Mark holds a double B.Sc. in Physics and Electrical Engineering from the Technion and an M.Sc. in Neuroscience from the Gonda Multidisciplinary Brain Research Center at Bar-Ilan University.
Quantization at Hailo – Co-Evolution of Hardware, Software and Algorithms
“Power-efficient hardware for deep neural network (DNN) acceleration is a major obstacle for the successful deployment at scale of many edge AI applications such as autonomous driving, smart retail, smart cities and more. Quantization is exploited by modern DNN acceleration at the edge to significantly reduce power consumption. In the first part of the talk, we will give a brief overview of the hardware aspects of quantization, followed by a high-level review of the main approaches to quantization. We show how these concepts are leveraged in the Hailo-8, a highly power-efficient DNN accelerator. In the second part, we discuss "real-world" challenges of quantization as well as suggest perspectives for future work to address current gaps.”
Datagen Technologies Ltd.
Dr. Jonathan Laserson is the Head of AI Research at Datagen, and an early adopter of deep learning algorithms. He did his
bachelor studies at the Israel Institute of Technology, and has a PhD from the Computer Science AI lab at Stanford University.
After a few years at Google, he ventured into the startup world, and has been involved in many practical applications of ML and
deep learning. Most recently, at Zebra Medical vision, he led the development of two FDA-approved clinical products, applying
neural networks to millions of medical images and unstructured textual reports.
Embedding Synthetic Assets Using a Neural Radiance Field
At Datagen, we maintain a large catalogue of artist-made 3D assets from multiple object categories (i.e. tables, chairs, plates,
bottles, etc.). Each asset object consists of a detailed 3D mesh and a texture map. Searching through these objects (for example,
in order to find "all tables with 3 legs") can only be done using their meta-data attributes, which were entered manually by the
artists, often without the desired attributes. We instead propose to encode each object in an embedding space, which encapsulates
the entire object's shape and appearance, and thus allows extracting any desired feature from it, without the need to render the
object. We use a single scene representation network (SRN) to form an implicit volumetric representation of all objects in a specific
category, where the shape and appearance are decoupled. Combined with a differentiable renderer, the network is trained to
ensure that images rendered from the SRN conditioned on an object's code, are similar to the images rendered directly from that
object's textured mesh. Moreover, by manipulating the latent vectors, one can generate new objects, or combine shape and
appearance from multiple objects. Examination of the induced embedding space confirms that the assets are organized in a way
that permits visual ad-hoc classification tasks, based only on their latent vectors.
PhD student for computer sciences at the TechnionTechnion
Yossi is a PhD student at the Technion, researching computer vision and planning algorithms for robotic systems. He holds an M.Sc in Physics from Tel-Aviv University and B.Sc in Electrical Engineering and B.Sc in Physics From the Technion.
Local Trajectory Planning For UAV Autonomous Landing
An important capability of autonomous Unmanned Aerial Vehicles (UAVs) is autonomous landing while avoiding collision with obstacles in the process. Such capability requires real-time local trajectory planning. Although trajectory-planning methods have been introduced for cases such as emergency landing, they have not been evaluated in real-life scenarios where only the surface of obstacles can be sensed and detected. We propose a novel optimization framework using a pre-planned global path and a priority map of the landing area. Several trajectory planning algorithms were implemented and evaluated in a simulator that includes a 3D urban environment, LiDAR-based obstacle-surface sensing and UAV guidance and dynamics. We show that using our proposed optimization criterion can successfully improve the landing-mission success probability while avoiding collisions with obstacles in real-time.
Graduate Student in the School of Computer Science Tel Aviv University
StyleCLIP: Text-Driven Manipulation of StyleGAN Imagery
StyleGAN is able to generate highly realistic images in a variety of domains, and therefore much recent work has focused on understanding how to use the latent spaces of StyleGAN to manipulate generated and real images. In this talk, I will present our recent paper "StyleCLIP: Text-Driven Manipulation of StyleGAN Imagery". In this paper, we explore leveraging the power of recently introduced CLIP models in order to develop a text-based interface for StyleGAN image manipulation. This interface provides a great expressivity for image editing and allows to perform edits that were not possible with previous approaches.
Computer Vision Algorithm Engineer at 3DFY.ai. Technion
Rajaei Khatib is a Computer Vision Algorithm Engineer at 3DFY.ai. Rajaei obtained both B.Sc. and M.Sc. from the department of Computer Science at the Technion, where he was advised by Prof. Michael Elad. His research focused on the connection between sparse representation and deep neural networks.
Learned Greedy Method (LGM): A novel neural architecture for sparse coding and beyond
The fields of signal and image processing have been deeply influenced by the introduction of deep neural networks. Despite their impressive success, the architectures used in these solutions come with no clear justification, being ‘‘black box’’ machines that lack interpretability. A constructive remedy to this drawback is a systematic design of networks by unfolding well-understood iterative algorithms. A popular representative of this approach is LISTA, evaluating sparse representations of processed signals. In this paper, we revisit this task and propose an unfolded version of a greedy pursuit algorithm for the same goal, this method is known as LGM.
Master's student for Physics Bar-Ilan University
Re'em Harel is a Master's student for Physics at Bar-Ilan University whilst working for the Israel Atomic Energy Commission (IAEC)
Complete Deep Computer Vision Methodology for Investigating Hydrodynamic Instabilities
In fluid dynamics, one of the most important research fields is hydrodynamic instabilities and their evolution in different flow regimes. Currently, three main methods are used for understanding such phenomena -- namely analytical and statistical models, experiments, and simulations -- and all of them are primarily investigated and correlated using human expertise. This presentation/work demonstrates how a major portion of this research effort could and should be analyzed using recent breakthrough advancements in the field of Computer Vision with Deep Learning. Specifically, Image Retrieval, Template Matching, Parameters Regression, and Spatiotemporal Prediction -- for the quantitative and qualitative benefits they provide. In order to do so, this research focuses mainly on one of the most representative instabilities, the Rayleigh-Taylor instability. The techniques which were developed and proved in this work can serve as essential tools for physicists in the field of hydrodynamics for investigating a variety of physical systems. Some of them can be easily applied to already existing simulation results, while others could be used via Transfer Learning to other instabilities research.
Applied Research Scientist at DAMO AcademyAlibaba Group
Emanuel Ben Baruch is an applied researcher at the Alibaba DAMO Academy, Machine intelligence Israel lab. His main fields of interests are deep learning approaches for image understanding as multi-label classification and object detection. Before joining Alibaba, Emanuel worked as a Computer Vision algorithm developer in Applied Materials and in an Israeli defense company.
Emanuel holds BSc and MSc Degrees in Electrical Engineering, Specializing in statistical signal processing, both from Bar Ilan University.
Asymmetric Loss For Multi-Label Classification
Pictures of everyday life are inherently multi-label in nature.
Hence, multi-label classification is commonly used to analyze
their content. In typical multi-label datasets, each picture contains only a few positive labels, and many negative ones. This
positive-negative imbalance can result in under-emphasizing
gradients from positive labels during training, leading to poor
In this lecture, we will introduce a novel asymmetric loss (”ASL”),
that operates differently on positive and negative samples.
The loss dynamically down-weights the importance of easy
negative samples, causing the optimization process to focus
more on the positive samples, and also enables to discard mislabeled negative samples.
We demonstrate how ASL leads to a more ”balanced” network, with increased average probabilities for positive samples, and show how this balanced network is translated to better mAP scores, compared to commonly used losses. Furthermore, we offer a method that can dynamically adjust the level
of asymmetry throughout the training.
With ASL, we reach new state-of-the-art results on three
common multi-label datasets, including achieving 86.6% on
MS-COCO. We also demonstrate ASL applicability for other
tasks such as fine-grain single-label classification and object
ASL is effective, easy to implement, and does not increase
the training time or complexity.
Code is availabl: https://github.com/Alibaba-MIIL/ASL
Phd. Candidate in Applied Mathematics, Research Scientist at eBayTel-Aviv University
Ido is a Phd. Candidate in Applied Mathematics at Tel-Aviv University, as well as a research scientist at eBay research. He is interested in representation learning, model interpretability, and theoretical deep learning. Before his work at eBay, Ido worked as a team leader at DeePathology.ai.
Sparsity-Probe: Analysis tool for Deep Learning Models
We propose a probe for the analysis of deep learning architectures that is based on machine learning and approximation theoretical principles. Given a deep learning architecture and a training set, during or after training, the Sparsity Probe allows to analyze the performance of intermediate layers by quantifying the geometrical features of representations of the training set. We talk about the importance of representation learning, and geometrical features in the latent space.
Graduate StudentTel-Aviv University
ReStyle: A Residual-Based StyleGAN Encoder via Iterative Refinement
PhD student at the applied Mathematics Program, Mathematics departmentTechnion
Samah is a Ph.D. candidate at the applied Mathematics department, working on her research under the supervision of Dr. Moti Freiman, from the Biomedical engineering faculty. Currently, she is conducting research on Bayesian Deep-learning methods for MRI Registration. She received B.Sc. and M.Sc. (Cum Laude), both from Technion's Viterbi faculty of electrical engineering in 2017 and 2019, respectively. She supervises image-processing-related projects. Previously, she worked at Intel and served as a TA in various courses and tutor in Landa (equal opportunities) project. Samah is a fellow of the Ariane de Rothschild Women Doctoral Program.
Non-Parametric Bayesian Deep-learning Method for Brain MRI Registration
Recently, deep neural networks (DNN) are successfully employed to predict the best deformation field by minimizing a dissimilarity function between the moving and target images. Bayesian DNN models enable safer utilization in medical imaging, improves generalization and assesses the uncertainty of the predictions. In our study, we propose a non-parametric Bayesian approach to estimate the uncertainty in DNN-based algorithms for MRI deformable registration. We demonstrated the added-value of our Bayesian registration framework on the brain MRI dataset. Further, we quantified the uncertainty of the registration and assessed its correlation with the out-of-distribution data.
ResearcherAlibaba Damo Academy
Dr. Yonathan Aflalo holds a MSc. from Ecole Polytechnique in France and a PhD from the Technion where he specialized in spectral analysis of geometrical shapes. He is currently working as a researcher in Alibaba Damo Academy, specialized in deep learning model optimization, including deep neural networks pruning, and network architecture search to discover very efficient models able to run in real time on any platform.
Realistic use of neural networks often requires adhering to multiple constraints on latency, energy and memory among others. In Alibaba, models deployed in production need to have a very low inference time to answer several cost constraints. In this talk we present a Neural Architecture Search (NAS) method that we have developed to construct these models by introducing Hard Constrained diffeRentiable NAS (HardCoRe-NAS), that is based on an accurate formulation of the expected resource requirement and a scalable search method that satisfies the hard constraint throughout the search. Our experiments show that HardCoRe-NAS generates state-of-the-art architectures, surpassing other NAS methods.
Algorithm ResearcherForesight Autonomous
Omri is a computer vision researcher at the R&D team of Foresight Autonomous.
His main research topics are sensor pose estimation and 3D reconstruction. He holds his BSc in Computer Science from Ben-Gurion University.
Multi Sensor Quality Assessment Using 3D Reconstruction
Many systems have sensor redundancy thus require real time assessment of the data quality and relevance from each source. The assessment is useful when allocating the computation resources and user interface. It is difficult to reliably make such comparison between different sensor types in real time as that usually requires data registration and/or deep understanding of the observed scene.
We present a method for assessing and comparing data quality and relevancy through 3D reconstruction, such as by stereo vision. This method allows data assessment to be performed in real-time before other complex algorithms. By using this approach, we can avoid confusing data from noisy channels, and concentrate our resources on the best channel.
Chief Scientist & Co-founder Brodmann17
Using Synthetic Data for ADAS Applications and Perception Challenges
The performance of the Artificial Intelligence (AI) that powers automotive systems is directly linked to the data used to train and test it. Our research tackles the issue of using synthetic data for two different Computer Vision tasks where relevant data is hard to collect or simply doesn’t exist. We show how to train object detectors, based only on synthetic data, that has generalized from real-world data. We then show how to use synthetic data to evaluate the performance of distance estimation algorithms. The success of these two tasks paves the way for future research taking advantage of synthetic data.
Open as PDF
94, Yigal Alon St.
Tel Aviv 6109202