- A state of the art review of Machine Learning & Deep Learning
TO AVOID CONFUSION - REGISTRATION IS ON EVENTBRITE - https://www.eventbrite.com/e/a-state-of-the-art-review-of-mldl-tickets-49041067152 Hello all, Our next event will be part of the awesome DLD 2018 conference where we are hosted by Samsung NEXT. The agenda for the Meetup includes: 16:00 - 16:20: Gathering 16:20 - 17:00: Dr. Guy Katz to speak on "Verification and Certification of Deep Neural Networks" 17:00 - 17:30 Deep Learning Research - A Startup Approach by Amir Alush, CTO @ Brodmann17 17:30 - 18:00: We don't need no labels: the future of pertaining and self-supervised learning, Bar Vinograd. TO AVOID CONFUSION - REGISTRATION IS ON EVENTBRITE - https://www.eventbrite.com/e/a-state-of-the-art-review-of-mldl-tickets-49041067152 Talks Abstracts: We don't need no labels: the future of pretraining and self-supervised learning: Telling a cat from a bird? that's easy, most infants can do that. But how about learning to paint a black and white photo with real colors? It takes years of practice and skill. It's time for your models to grow up. Most researchers and industry practitioners use pretrained models. We find that transfer learning from different datasets and tasks saves a lot of time and money when labels are scarce and data is limited. However, results are probably far from optimal. In this lecture, I will review self-supervised methods that are used to pretrain models on unlabeled data. I will cover techniques for Image, Video, Audio, NLP and other domains from recent literature (e.g. Colorization, Language Models, Super Resolution). These methods have been proven to show substantial gains over standard pretraining like those done with ImageNet. Finally, I will refine best practices to use and train self-supervised models so you can apply them to your data. Speakers Bios: Bar Vinograd: Bar is an expert consultant in the fields of machine learning and and deep learning. He has been working closely with startups, corporates and leading institutions for the past 5 years. His fields of expertise include Computer Vision, NLP, Statistical Modelling and Anomaly Detection. Bar also has 8 years of experience in software engineering and is a graduate of 8200 IDF Unit and of Tel Aviv University in Math and Linguistics. As an active community member, Bar is frequent at giving talks about machine learning, a mentor at Google AI launchpad and a teacher and an advisory board member at Israel Tech Challenge.
- The Race for AI: Lessons from building deep learning recommendation systems.
Agenda: 18:00 - 18:30 : Gathering 18:30 - 20:00 : The Race for AI: Lessons learned while building deep learning architectures for recommendation systems. Abstract: Deep Learning models have been gaining increasing attention in the recommendation systems community, replacing some of the traditional methods. The sparse nature of the problems in this domain and the different types of inputs offer unique challenges for feature engineering and architecture planning, in order to balance between memorization and generalization. During the past year the algorithms team in Taboola moved all of our algorithms to deep learning models and in this talk we will share the lessons we learned doing so. Specifically, we will talk about building neural networks with multiple input types (click history, text and pictures); feature engineering in the deep learning era; capturing interactions between features using both deep dense architectures and Factorization Machine models; Tradeoffs between deep models, shallow models and the combination of the two; and other tips regarding network architectures. Speaker Bio: Ofer Alper is an Algorithm Team Leader at Taboola, the world's largest discovery platform. In his current position he is responsible for building a machine learning based recommendation system, using neural network, deep learning and matrix factorization. Before joining Taboola Ofer worked in the Algo-Trading industry developing algorithms and data science models. Prior to that he was the R&D Manager of Mocospace, one of the largest mobile social networks in the US. Ofer has over 17 years of experience in many areas including software engineering, algorithm development, data science and machine learning.
- Interaction Based Feature Extraction: From Users’ Activity to Valuable Features
Agenda: 18:00 - 18:30 : Gathering 18:30 - 20:00 : Interaction Based Feature Extraction: How to Convert Your Users’ Activity into Valuable Features. Today almost every website and app collect data about the interactions (clicks, likes, views...) between users and items. The most common use case for these sparse “user-item” matrices is to train and improve different recommendation systems. In my presentation I will introduce how we can use exactly the same matrices together with additional datasets to generate valuable features that can be used to train different regression and classification models. I will start with describing how it was implemented at SimilarWeb, in order to accurately estimate different website metrics like demographics (age and gender) and category and continue with explaining how we can expand the algorithm to solve similar problems in different domains. Speaker: Shlomi Babluki (https://il.linkedin.com/in/shlomi-babluki-1a29289 ) , Data Analysis Team Leader, SimilarWeb.
- The Race for AI: Model Serving for Deep Learning
Agenda: 18:00 - 18:30 : Gathering 18:30 - 20:00 : The Race for AI: Model Serving for Deep Learning Abstract: Deep Learning has been delivering state of the art results across a growing number of problems and domains. Correspondingly, Deep Learning models are being deployed across a growing number of applications and use cases. In this talk, you will learn about what deploying deep neural networks to production mean, design considerations and challenges for model serving, and how the open source project Model Serving for Apache MXNet is designed to address these challenges. Speaker: Hagay Lupesko (https://www.linkedin.com/in/hagaylupesko/) is a Deep Learning engineering leader at AWS, he will share his knowledge and experience in the field.
- A-Z Deep Learning based object detection
• What we'll do Object Detection might be one of the most important tasks of Computer Vision. In the last few years, Deep Learning has rocketed the accuracy of object detection algorithms. In this meetup, Assaf Mushinsky will teach you all there is to it about object detection. From the most basic concepts and algorithms to the most advanced and recent methods. This meetup will be for both beginners and experts. No prior knowledge of Object Detection is needed but experience with machine/deep learning is required. This meetup will be composed out of the following four parts: 1. Introduction to object detection, evaluation metrics and the history of object detection. 2. The basics of deep learning based object detection methods, starting from sliding window approaches to region based methods. 3. Faster solutions for object detection such as Faster R-CNN and SSD. 4. Review of more recent deep learning object detection methods. About the speaker: Assaf Mushinsky is the Co-founder and Chief Scientist of Brodmann17. He is an expert in Deep Learning and computer vision with vast industry experience. He hold a MSc. from Tel-Aviv university under the guidance of Prof Lior Wolf. Some notes: -The lecture will be given in English -This meetup is co-hosted with "Living on the edge" meetup group • What to bring • Important to know
- Data Science at Scale - Lessons from Stitch Fix
Agenda: 18:30-19:00 - Gathering (food and drinks) 19:00 –20:00 - Data Science at Scale - Lessons from Stitch Fix (https://www.stitchfix.com/) by Randy Shoup, VP Engineering at Stitch Fix Talk Abstract: Stitch Fix is a clothing retailer in the United States, and we use data science in all aspects of our business to help our clients find the clothes they love. We leverage the combination of machine learned models with human expert decision making across functions from buying to logistics to selection to shipping. And because Stitch Fix sells over USD1 billion in clothes each year, we do all this at scale. This session will cover three parts of scaling data science: * Scaling Data - how we collect, aggregate, and process data for our machine-learned models * Scaling Data Scientists - how we provide a simple but powerful Platform-as-a-Service tailored to the needs of our data scientists * Scaling Learning - how we run experiments to refine and improve our algorithms The Bio : Randy Shoup is the VP of Engineering at Stitch Fix. Randy is a 25-year veteran of Silicon Valley, and has been providing tools and infrastructure to data scientists, and has been building and operating applications using machine-learned data, for over a decade. As Chief Engineer at eBay, he helped to introduce machine-learned ranking to the search engine. And at Stitch Fix, his teams leverage data science every day. He is a frequent speaker on topics ranging from engineering organization to scalability to DevOps. • Important to know Gathering, eating and mingling starts at 18:30, talk starts at 19:00
- The Race for AI: Deep Learning - Past, Present & Future
• What we'll do Agenda: 18:00-18:30 - Gathering (food and drinks) 18:30 –20:00 - The Race for AI: Deep Learning - Past, Present & Future by Eran Paz, CTO @ VizScribe. Talk Abstract: Although deep learning has been around for over 20 years, it gained most of its popularity only in recent years. The emergence of Alexnet late in 2012 is considered ancient history in this fast paced field and new applications and architectures are published constantly. In this talk we will review some basics deep learning architectures, take a look into more recent and advanced models and peek into the hottest topics in the field today This is an introduction level, non technical talk and doesn't require any prior knowledge. Speaker Bio: Eran is a deep learning researcher at SAP innovation center, where he's working on deep learning research for almost 4 years. His main research areas are object detection in videos, tracking, generative models and more. Eran hold's a B.Sc and M.Sc in Industrial engineering from Ben Gurion Univ and working towards finishing his M.Sc in computer science from the Open University. • Important to know Gathering, eating and mingling starts at 18:00, talk starts at 18:30
- The Race For AI: An Historical Overview of Semantic Image Segmentation
Agenda: 18:00-18:30 - Gathering (food and drinks) 18:30 –20:00 - The Race For AI: Semantic Image Segmentation - Historical Overview by Shai Harel, CEO @ VizScribe. Talk Abstract: In this session we will cover the history of semantic image segmentation, from old colour based method, to edge based/ super pixel method, and finally go trough the recent revolutions in this area such as CRF and GAN. No prior knowledge is required, the idea is to give the intuition behind the equations. Speaker Bio: Over 15 years of experience in computer programming, 10 years in the field of statistical machine learning. Finished my M.Sc at the open university of Israel in the field of computer vision, under the supervision of Prof. Tal Hassner. Led the computer vision effort @corrigon, which has been acquired by eBay. Currently acting as the CEO of VizScribe. Expert in field of 2D and 3D alignment and deep learning.
- The Race For AI: Interpreting Sarcasm with Sentiment Based Machine Translation
Agenda: 18:00-18:30: Gathering (food and drinks) 18:30 –20:00- "The Race For AI: Interpreting Sarcasm with Sentiment Based Machine Translation" By Lotem Peled from Gong.io Talk abstract: Sarcasm is defined as the use of words that mean the opposite of what one would really want to say in order to insult someone, to show irritation, or to be funny. In other words, "Sarcasm - the giant chasm between what I say, and the idiot who doesn’t get it". From Facebook and Twitter to IMDB and Amazon reviews, sarcasm is frequently found in opinionated user generated content, posing a challenge both to human readers and to NLP algorithms (such as sentiment analysis). Machines can nowadays identify sarcastic expressions; but other than identifying the sarcasm, can they actually understand and interpret it? In this talk, I will present my research on the novel task of sarcasm interpretation, along with its motivations and challenges. I will present a parallel corpus of sarcastic tweets and their non sarcastic interpretations constructed for this task, and introduce the Sarcasm SIGN - a machine translation based sarcasm interpretation algorithm that puts a special emphasis on sentiment words. I will conclude with a discussion on how to evaluate the quality of generated sarcasm interpretations, and present directions for future research on this new task. Speaker bio: Lotem Peled is a data scientist and NLP researcher at Gong.io (http://gong.io/), with an MSc in Natural Language Processing from the Technion. Working at Gong.io (http://gong.io/) Lotem specializes in neural language modeling, while her academic research focuses on the task of sarcasm interpretation, its evaluation, and applications in multiple domains. Lotem's latest work, presenting a novel system for sarcasm interpretation, has been presented in ACL 2017, the world's leading NLP venue.
- DLD 2017 Meetup - “Modern Race to AI and Deep Learning”
This Meetup is a collaboration between AI IL and Big Thing. Agenda: 18:00 - 18:30 - Mingling 18:30 - 19:15 - The mechanics of Deep Learning - the Engineer version - Yosi Taguri @ Missing Link AI 19:15 - 20:00 - The Race for AI: Deep Reinforcement Learning - Yam Peleg @ Deep Trading Title: The mechanics of Deep Learning - the Engineer version - Yosi Taguri @ Missing Link AI Abstract: Deep Learning is a super power for engineers who master it. It allows to solve complex and abstract problems that were the domain of Data Scientist and PhD's who'd spent years in academia. In this session we'll understand the mechanics behind Deep Learning and what makes them so powerful. By the end of this session you will be able to build your own Neural Network and start to explore a super power you didn't know you have. Bio: A developer since the age of 11 when computers had 8 bits and less than 64k of memory. Passionate about People and Technology. For the past 25 years he has developed software on a wide range of technologies and devices. He is a frequent speaker on several events ranging from Marketing to Technical conferences. Started developing software at the age of 11 and sold his first program at the age of 14. Title: The Race for AI: Deep Reinforcement Learning - Yam Peleg @ Deep Trading Abstract: You may have noticed that computers can now automatically learn to play ATARI games and they are beating world champions at Go. All of these advances and some of the future ones will fall under the umbrella of Deep Reinforcement Learning, The art of training and AI model based on positive and negative reinforcements. In this tutorial we will review and learn about Deep Reinforcement Learning, from the pioneering research of the fields to hands on demos live. Bio: Yam Peleg is the founder of Deep Trading ltd. He was a quantitative high frequency trader for more than four years and deep trading is his second startup. He is also a major contributor to the python community and one of the contributors of Keras - the leading deep learning code base, who spoke at python conferences around the world, including PyCon, PyData ,SciPy and more.