- Visualizing neural network activity
In this session, Justin is going to examine feature salience, and try to open the black box of deep learning algorithms. For example, visual features identified in image classification tasks are inherently suitable for representing the decision boundary of a classifier. But what to do when the features are audio-based? Or syntactic? Join us for an overview of current innovative efforts to expose the selection process of deep learners and a lively discussion on how to communicate the inner workings of your favorite deep learning algorithm. About the speaker: Justin is a research assistant in the Institute of Mathematics at the University of Osnabrück, Germany. He is also an Intel software innovator from the US and studies cognitive science in Germany with a focus on machine learning and computer vision. Before studying cognitive science, he worked in neuroscience research with neuronal and brain activation imaging. He founded and organized the San Antonio Science Café and the Tel Aviv Science Café, and is preparing his master's thesis on visualizing deep learning neural networks.
- Building an AI platform
Welcome everyone, On Wednesday, we are going to have a talk by Amir Khosrowshahi. Amir is going to present Intel’s strategy to provide deep learning as a core technology that is easy to use, high performance, scalable, and power efficient. This involves combining a diverse portfolio of hardware architectures with layers of software, services, and machine learning algorithms. We will discuss challenges encountered and how they were addressed. About the speaker Amir is a VP and CTO of the AI products group at Intel. He was co-founder and CTO of Nervana, a startup providing a full-stack deep learning platform. Nervana was acquired by Intel almost a year ago. He has a research background in neuroscience, machine learning, and distributed computing. He started his career at Goldman Sachs where he was a derivatives trader. Amir has a PhD from Berkeley and AM and AB from Harvard.
- Deep Learning for speech recognition & Compressing neural nets for IoT devices
Speech Recognition: Speech recognition is invading our lives. It’s built into our phones, our game consoles and our smart watches. It’s even automating our homes. But speech recognition has been around for decades, so why is it just now hitting the mainstream? In the first session we will show how deep learning finally made speech recognition accurate enough to be useful outside of carefully controlled environments. Compressing Neural Nets: We have reached a period in which Computer Vision tasks are almost solved using Deep-Learning techniques but some open questions still remain around practical usage of these techniques, specifically in IoT applications. In the second part of the lecture we will cover how to compress Deep neural nets to fit IoT edge devices. Speaker Bio: Alan Bekker is a Phd researcher and an author of papers in leading artificial intelligence journals and conferences, mostly focusing on applications of deep learning. https://alanbekker.wordpress.com/ Target Audience: Some basic deep learning background is needed
- AWS Deep Learning with MXNet
Dear friends, You might be interested in this event tomorrow. Please don't register here, register at the original meetup page: AWS Deep Learning with MXNet https://www.meetup.com/Big-Data-Israel/events/236382771/?_af=event&_af_eid=236382771
- Fast Deep Learning at Your Fingertips
Speaker: Dr. Amitai Armon, Chief Data Scientist, Intel Advanced Analytics Abstract: Deep learning is considered the most significant innovation in data science in recent years, and it presents amazing improvements in the modeling results. However, most data scientists don’t use deep learning yet, for several reasons: the relative complexity of customizing deep learning models for their own problems, challenges in installing and using the required frameworks, and low performance of open source deep learning frameworks on standard CPUs. We present a new, free software tool that enables the creation of deep learning models quickly and easily. The tool is based on existing deep learning frameworks and incorporates extensive optimizations that provide high performance on standard CPUs. Target audience: The talk is intended for data-scientists who are interested to start using deep-learning, as well as those who already use it and would like to learn about this new CPU-based tool.
- The Return of The Neural Network: Zeroing-in on the Human Brain.
Target audience: this is an intro talk for developers. Lecturer: Dr. Jonathan Laserson is a senior algorithm researcher at PointGrab and a Machine Learning expert and consultant. He has a PhD from the Computer Science AI lab at Stanford University and was a lecturer at Bar-Ilan University. After 3 years doing machine learning at Google, today he is focused on Deep Learning algorithms and their practical usage on embedded architectures. Abstract: Neural networks, born in the 40s, revived in the 80s, and neglected in the 90s, are now making their second Artificial Intelligence comeback, and this time they are winning. Coined by the much cooler name Deep Learning, these algorithms are beating classical Machine Learning algorithms by a significant margin in scores of AI challenges. All the while, Deep Learning theoreticians are lagging behind, still having no clue how come it works so well. And the best thing still - these algorithms are pretty simple to understand, and the software infrastructure available to run them at scale is outstanding and supported by a huge developers community. In the talk I will give an introductory talk about Deep Learning and how to get started with it.
- Deep Learning for Facial Recognition
Abstract: In Parallel to other tasks in the field of Computer Vision, the popular task of face recognition achieved outstanding results using the emerging Deep Learning technology. Facebook's "DeepFace" paper was the first to describe the remarkable results of near human level recognition. However, these results were hard to reproduce without the massive datasets available to the Facebook AI team and its partners. In recent years, many new facial datasets have been collected and made available for research studies. Several of these studies show improvements that managed to outperform human levels of recognition on common datasets and describe newly introduced applications, such as gender classification, age estimation and facial attributes detection. In the first talk, Yaron will review key papers in the field of face recognition, will analyze the most significant state-of-the-art approaches and will present recent facial datasets and challenges. In the second talk, Yair will review recent papers on the newly popular task of facial attributes and discuss their possible applications. About FDNA: FDNA has developed Face2Gene, a free genetic search and reference tool powered by the Facial Dysmorphology Novel Analysis (FDNA) Technology. Face2Gene facilitates detection of dysmorphic facial features and recognizable patterns of human malformations from a facial photo to present a list of matching syndromes with comprehensive and up-to-date references. Speakers: Yaron Gurovich, VP R&D & Site manager, FDNA Inc. Yair Hanani, Algorithms group manager, FDNA Inc. Yaron and Yair are experienced Computer Vision and Machine Learning developers. They both hold MSc degrees in Computer Vision from TAU under the supervision of Prof. Lior Wolf. Today Yaron is the VP R&D at FDNA and Yair is the Algorithms group manager.
- Deep Learning in Finance
Abstract: Deep learning has recently gained considerable attention in the speech transcription and image recognition community for its superior predictive properties. However, its application to financial market prediction has not been heavily researched yet. This talk is a high level presentation of deep learning techniques used in finance and a review of various papers published on the topic. In addition Yam will share some of his personal trading experience using deep learning and classical machine learning. Speaker 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 who spoke at python conferences around the world, including PyGotham, PyData and SciPy. He started his computer science degree studies at the age of 14 and throughout the years he had been using his knowledge to give services of cyber security consulting, penetration testing and vulnerability researching and he stands behind the publication of some major security leaks in companies around the world. Target Audience: Researchers in the fields of machine learning and deep learning. Basic knowledge of deep learning required.
- A Shallow Introduction to Deep Learning
This meetup is sponsored by Satellogic (http://www.satellogic.com/)! Abstract: I will give a short introduction to deep learning with historical perspective. Most of the talk will be dedicated to showing the latest examples in image processing, text understanding, and other applications. I will conclude with pointers to architectures, frameworks, hardware and references. Target audience: This is an introductory non-technical talk intended for non-pros (programmers or researchers not working with deep learning). Bio: Dr. Eyal Gruss is a machine learning researcher and consultant, with experience in algotrading, cybersecurity, chemometry, adtech and NLP. Eyal is a Talpiot graduate and did his PhD in theoretical physics. He is also a digital artist, creating interactive installations and computer generated art. Sponsor: In 2016, Satellogic will be launching its constellation of mass-produced imaging satellites, and begin to collect the largest up-to date dataset of high-resolution images of the planet. We are looking for Deep Learning researchers and practitioners to help us derive value from this data and disrupt the industry by transforming space-based imaging and data into an accessible part of responsible daily decision-making for any business.
- Oded Luria: Pricing Illiquid Assets using Deep Learning
Abstract: Deep Learning has been shown to surpass conventional methods in many learning tasks such as image and voice recognition, but its role in processing financial datasets has yet to be fully discovered. In this Meetup, Oded will share our group experience in applying Deep Learning to price financial assets. Oded will present some of the unique challenges and tradeoffs of this field, discuss relevant aspects of using Deep Learning in the financial industry and demonstrate how Deep Learning can be applied to improve existing pricing algorithms. Target audience: beginner to high-level DL practitioners, especially for those who are interested in applied research aspects of Deep Learning. Oded is a Senior Data Scientist in the Citi Innovation Lab. Before joining Citi, Oded worked as a Senior Quantitative Researcher for a large international hedge fund and as an Algorithm Engineer for several start-up companies. Oded earned his bachelor degree in Electrical Engineering, and his master and Ph.D. degrees in Biomedical Engineering, all from Tel Aviv University.