• DLBLR Monthly Meetup - June 2019

    Intuit

    Productionizing deep learning workflow with Hangar, PyTorch/Tensorflow & RedisAI • What we'll do 9:00 am: Doors open and networking 9:30 AM: Deep Learning workflow 10:00 AM: Introduction to Hangar 10:30 AM: Introduction to RedisAI 11:00 AM: Walkthrough - PyTorch & Tensorflow 11:30 PM: Training in Colab, deploying in the cloud PS: We'll have to accommodate 2 to 3 breaks in between these sessions and hence we are on a tight schedule. So please try not to be late if not early. Historically, we have learnt that the initial and final networking sessions are way more productive than we think. Please make use of that as well. This meetup is always on a first-come-first-serve basis. The auditorium can accommodate about 100 participants. The meetup is not live streamed unless and otherwise specified. Venue: Intuit office, Pritech Park, *Building 8*, RMZ Ecospace Internal Rd, RMZ Ecospace, Adarsh Palm Retreat, Bellandur, Bengaluru[masked]https://goo.gl/maps/YQfDtz5wYCNRwyF28

    8
  • DLBLR Monthly Meetup - May 2019

    Intuit

    • What we'll do This meetup is always on first come first serve basis. The auditorium can accommodate 70-80 participants. The meetup is not live streamed unless and otherwise specified. The meetup typically has two sessions, each with 75 minutes with 15 minutes of break between two sessions. 9:00 am: Doors open and networking • Session 1: 9:30 to 10:45am Arjun and team, Agara Labs Title: Deep Learning - view from trenches Abstract: The talk will consist of two parts. The first part covers building text-based deep learning systems from scratch to production-level. Topics touched upon will include but not be limited to : the use of attention mechanisms, multi-task learning, working with embedding spaces, and custom loss functions. Practical implications of building such models such as maintainability, cost, etc will also be discussed. The second part covers some fundamental challenges and approaches to speech-based deep learning systems. Topics discussed in this part will include but not be limited to analysing and enhancing the outputs of state-of-art Automatic Speech Recognition systems (speech-to-text models), and practical implications. • Session 2: 11:00 am to 12:15 pm Madhu Gopinathan, MMT Title: Quickly creating training data to build deep learning models for natural language processing Abstract: Large labelled training sets are essential to power deep learning models for natural language processing applications. Creating labelled data is an expensive, error prone and time consuming process, which slows down the development of such applications. In this talk, we will discuss recent research – a paradigm called data programming - and an implementation: Snorkel to illustrate how this approach can aid in analysing natural language data and creating labels through programming. Reference: Data Programming: Creating Large Training Sets, Quickly, Ratner et. al., NeurIPS 2016 https://arxiv.org/abs/1605.07723 Venue: Intuit office, Pritech Park, *Building 8*, RMZ Ecospace Internal Rd, RMZ Ecospace, Adarsh Palm Retreat, Bellandur, Bengaluru[masked]https://goo.gl/maps/YQfDtz5wYCNRwyF28

    6
  • DLBLR Monthly Meetup - June 2018

    NVIDIA Graphics Pvt. Ltd.

    • What we'll do This meetup is always on first come first serve basis. The auditorium can accommodate 80 to 100 participants. The meetup is not live streamed unless and otherwise specified. The meetup typically has two sessions, each with 75 minutes with 15 minutes of break between two sessions. • Topic: 'Deep learning for Graphs' • Speaker: Sushravya GM, researcher at Accenture AI Labs, Bangalore • Agenda 930am: Doors open and networking 10 to 11.15am - Deep learning for Graphs - session 1 11:30 am to 01:00 pm - Deep learning for Graphs - session 2

  • DLBLR Monthly Meetup - March 2018

    NVIDIA Graphics Pvt. Ltd.

    • What we'll do This meetup is always on first come first serve basis. The auditorium can accommodate 80 to 100 participants. The meetup is not live streamed unless and otherwise specified. The meetup typically has two sessions, each with 75 minutes with 15 minutes of break between two sessions. • Topic: 'Deep Learning with PyTorch' • Speaker: Sherin Thomas • Agenda 930am: Doors open and networking 10 to 11.15am - What is PyTorch - Building a deep net with PyTorch - Debugging and profiling 11:30 am to 01:00 pm - Just-in-time compilation - PyTorch to production and mobile devices - Interop with other frameworks with ONNX and visualization PyTorch is a year old DeepLearning framework introduced with dynamic graph capability in mind. PyTorch plays well with python and provide super intuitive and flexible functional APIs which is much easier to follow than almost any other DL frameworks. We will discuss the philosophy of PyTorch, the way it is built, and the internals. We will create deep networks and discuss how it can be moved to production/mobile devices. • Disclaimer: The session is, by no means, trying to prove PyTorch is better than any other frameworks. In fact, we'll see how PyTorch helps the research community to prototype the idea faster and move it to any framework of your comfort when you need it. • Prerequisite The audience is expected to go through the basics of Pytorch. I'll build the course from where I left in the last session I had given @anthill. My talk at anthill : https://www.youtube.com/watch?v=VMcRWYEKmhw • Additional resources: https://towardsdatascience.com/pytorch-tutorial-distilled-95ce8781a89c https://medium.com/init27-labs/pytorch-basics-in-4-minutes-c7814fa5f03d https://towardsdatascience.com/pytorch-vs-tensorflow-spotting-the-difference-25c75777377b

    2
  • DLBLR Monthly Meetup - February 2018

    NVIDIA Graphics Pvt. Ltd.

    • What we'll do This meetup is always on first come first serve basis. The auditorium can accommodate 80 to 100 participants. The meetup is not live streamed unless and otherwise specified. The meetup typically has two sessions, each with 75 minutes with 15 minutes of break between two sessions. The material during meetup may not be available publicly as the speakers may use proprietary content during their presentations. Participants can directly contact speakers for the material if the organizers don't make the material open to all. 930am: Doors open and networking 10 to 11.30am Zero to One with AlphaGoZero We will go through the basics of Q learning, Policy Gradient, Monte Carlo Tree Search, and then learn how AlphaGo team applied self-play learning to train a Go Player without ever seeing games played by human Go players. Speaker: Vijay Gabale No need to carry laptops. Useful reading https://web.stanford.edu/~surag/posts/alphazero.html https://medium.com/emergent-future/simple-reinforcement-learning-with-tensorflow-part-0-q-learning-with-tables-and-neural-networks-d195264329d0

    9
  • DLBLR Monthly Meetup - November 2017

    NVIDIA, Building L6, 8th Floor,

    This meetup is always on first come first serve basis. The auditorium can accommodate 80 to 100 participants. The meetup is not live streamed unless and otherwise specified. The meetup typically has two sessions, each with 75 minutes with 15 minutes of break between two sessions. The material during meetup may not be available publicly as the speakers may use proprietary content during their presentations. Participants can directly contact speakers for the material if the organizers don't make the material open to all. 930am: Doors open and networking We are looking for excellent speakers who can talk about recent advances in deep learning or end-to-end application of deep learning with technical depth as the first-class citizen. 10 to 11.30am This talk concerns with deep reinforcement learning and more specifically Deep Q-networks (DQN). DQN is first introduced by Google's DeepMind in 2014 for building general AI agents successful at playing Atari 2600 games at expert human levels [1]. The talk introduces reinforcement learning algorithms for building agents that operate in uncertain environments (model-free agents). Prerequisites: The audience are expected to have an some conceptual understanding of probability, machine learning including neural networks. However, to a beginner, it is a great opportunity to explore the topic; so beginners are encouraged too. References: [1] Mnih, Volodymyr; et al. (2015). "Human-level control through deep reinforcement learning" About the speaker -- Rahul is an independent software consultant and founder of an AI technology startup, Art Of Programming Edulabs. He works on challenging data-driven problems in education among other things. He is also a part-time trainer and consults technology startups on machine learning and software engineering.

    10
  • DLBLR Monthly Meetup - September 2017

    Location visible to members

    This meetup is always on first come first serve basis. The auditorium can accommodate 80 to 100 participants. The meetup is not live streamed unless and otherwise specified. The meetup typically has two sessions, each with 75 minutes with 15 minutes of break between two sessions. The material during meetup may not be available publicly as the speakers may use proprietary content during their presentations. Participants can directly contact speakers for the material if the organizers don't make the material open to all. 930am: Doors open and networking 10:00 am to 11:30 am: Game of Thrones approach to Deep Reinforcement Learning -Jaley Dholakiya, Data Scientist Harman Abstract: If you are watching game of thrones, then reinforcement learning should be peanuts for you. In this meetup, you need to have some background of Deep Learning and Game of Thrones before attending. There will be live code demonstration. We will be covering concepts leading upto Deep Reinforcement Learning via Game of Thrones inspired code demonstrations.

    9
  • DLBLR Monthly Meetup - June 2017

    NVIDIA Graphics Pvt. Ltd.

    This meetup is always on first come first serve basis. The auditorium can accommodate 80 to 100 participants. The meetup is not live streamed unless and otherwise specified. The meetup typically has two sessions, each with 75 minutes with 15 minutes of break between two sessions. The material during meetup may not be available publicly as the speakers may use proprietary content during their presentations. Participants can directly contact speakers for the material if the organizers don't make the material open to all. 930am: Doors open and networking 10am to 1115am: Visualizing, Understanding and Playing with ConvNets - by Arjun Jain Abstract: Deep learning models are often noted as “black boxes” in reference to the difficulties of tracing a prediction back to important features to understand how an output was arrived at. Although deep learning models are giving increasingly advanced results in diverse problems, their lack of interpretability is a major problem. However, there are a few techniques that can be utilized to get a peek into the workings of neural network. In this talk, we will look at algorithms with which we can gain some insights into the inner workings of our networks such as visualizing patches that maximally activate neurons, visualize the representation space, occlusion experiments, deconvolution approaches and optimization over image approaches. We will also look at DeepDream and NeuralStyle - algorithms with which we can generate impressive art from our images using ConvNets and have fun! Brief Bio: Arjun is the co-founder of Perceptive Code LLC and an Adjunct Assistant Professor at IIT Bombay. Prior to this, he was a researcher with a special project team at Apple in Cupertino. Before that he was also a post-doctoral researcher at the Computer Science department at New York University's Courant Institute where he worked with Yann LeCun. He received his Ph.D. in Computer Science from the Max-Planck Institute for Informatics in Germany. Broadly, his research lies at the interface of computer graphics, computer vision, and machine learning, with a focus on human pose estimation and data-driven artistic content creation tools. Arjun has worked as a developer for several companies, including Yahoo! in Bangalore and Weta Digital in New Zealand. He has been especially active in the visual effects industry. Arjun served as a developer for Weta Digital’s motion capture system. This system has been used in many feature films, and Arjun was credited for his work in Steven Spielberg’s, The Adventures of Tintin. Arjun’s work has resulted in several academic publications, a patent, and has been featured by mainstream media, including in the magazines: New Scientist, Discovery, BCC, Vogue, Wired, India Today, and The Hollywood Reporter, among other outlets. 1130am to 1245pm : TBA

    12
  • DLBLR Monthly Meetup - May 2017

    NVIDIA Graphics Pvt. Ltd.

    This meetup is always on first come first serve basis. The auditorium can accommodate 80 to 100 participants. The meetup is not live streamed unless and otherwise specified. The meetup typically has two sessions, each with 75 minutes with 15 minutes of break between two sessions. The material during meetup may not be available publicly as the speakers may use proprietary content during their presentations. Participants can directly contact speakers for the material if the organizers don't make the material open to all. This will be a two session (each session of 75 minutes) tutorial on deep learning with NLP. The first part will cover material for beginners while the second part will take deep dive into advanced techniques. The speaker also intends to take your through some of the code he had written (note: this is not a coding workshop). Abstract: An integral part of building any NLP applications is: building a rich representation of text. In past couple of years, there has been some phenomenal work in this area. In this meetup, we will take a deep dive into the world of word vectors. Starting from simplest models, we will journey through key results and ideas in this area. This will be a very hands-on session, covering - concepts, maths, and code. Speaker's Bio: Anuj Gupta is a senior ML researcher at Freshdesk, Chennai; working in the area NLP, Deep learning. He dropped out Phd in ML to work with startups. He graduated from IIIT H with specialization in theoretical comp science. Satyam Saxena is a ML researcher at Freshdesk, Chennai. An IIIT Jodhpur alumnus, his interest lie in NLP, Deep Learning. Prior to this, he was a part of ML group @Cisco and visiting researcher at Vision Labs in IIIT Hyd.

    15
  • DLBLR Monthly Meetup - March 2017

    NVIDIA Graphics Pvt. Ltd.

    **Agenda** Please carry a Government Photo ID. Visitor vehicle parking available at the lower basement. There is no fee to attend the meetup. We have limited (60 to 80) seats at the venue filled in First Come First Serve basis. The meetup will be open for sign up until it starts since not all those who sign up turn up at the meetup. We may or may not record or live stream the talks. If we do, we will post a link to the recording. Please don't assume recording or live streaming availability implicitly. 09:30 to 10:00 - Doors open, recap and networking. 10:00 - 11:15 - Title: Decoding Convolutional Neural Architectures Abstract: Convolutional neural architectures lie at the heart of the deep learning revolution witnessed over the past few years. The talk will trace the history of several convolutional neural architectures proposed for image, text and speech tasks. We will try to compare and understand how successive architectures try to overcome deficiencies of the earlier ones and incorporate new architectural patterns to improve performance. Bio: Nishant Sinha is a Computer Science Researcher, Advisor and Mentor at Kena Labs. He is interested in designing AI systems spanning textual, vision and speech modes, based on deep learning and logical reasoning. Earlier at IBM Research, he architected a human-machine conversation system to enhance collaborative decision making in business environments. Previously at NEC Labs, he developed intelligent inference tools for software error diagnosis at industrial scale. His current focus is to advise companies and mentor individuals on deep learning technology and solutions. Nishant obtained his Ph.D in Symbolic Reasoning from Carnegie Mellon University and B. Tech. in Computer Science from IIT Kharagpur. 11:15 to 11:30 - Short coffee break 11:30 to 12:45 Can neural networks learn to learn? Meta learning with deep learning Speaker : Apurva Gupta, Bloomreach

    18