• Special Topics 2: Applications of Deep Learning

    Redhorse Corporation

    Topics: Reinforcement Learning and LeNet5 Neural Network architecture. **** PRE-REGISTER TO PROVIDE MORE INFO HERE: https://forms.gle/4Gc6SamK5thqB2Pk6 Please complete this simple form to speed up the check-in process and to receive a slack channel invite. Wait listed attendees are encouraged to pre-register also. Our topics for this meeting will include a discussion on Reinforcement Learning led by Mark Shiffer. This is a continuation of last meeting's presentation and newcomers are welcome to jump in. The second half of our meeting will be a discussion of the LeNet5 Convolutional Neural Network architecture and examining implementations of LeNet5 in Python. Attendees are encouraged to try to implement a LeNet5 network beforehand to share what they have done and be able to ask/answer group questions. We have a slack channel setup for the folks who pre-registered and want to share information before/after the meetup. There are many resources on the internet but these will get you started. We encourage everyone to try whatever approach works best for you. LeNet-5 [original 1998, paper by LeCun et al.] http://yann.lecun.com/exdb/publis/pdf/lecun-01a.pdf High-level summary of LeNet-5. https://medium.com/@pechyonkin/key-deep-learning-architectures-lenet-5-6fc3c59e6f4 LeNet-5 in 9 lines of code using Keras https://medium.com/@mgazar/lenet-5-in-9-lines-of-code-using-keras-ac99294c8086 Implementing LeNet on MNIST dataset with Tensorflow https://www.kaggle.com/dwarkanath/implementing-lenet-on-mnist-dataset Special Topic meetings will occur about every 6 weeks. ------------------------------------------------------------ Our excellent host is Redhorse Corporation's Virginia offices, a short walk from the Rosslyn metro station. We thank Redhorse Corporation for their help and support in providing us a place to meet so we can continue to learn and share ideas. A couple of logistics items: 1) Feel free to arrive at 5:30 pm to allow for some time to get to the 12th floor and mingle with colleagues. We will begin the meeting at 6:00 pm. 2) Be sure to bring a photo ID. 3) To make the check-in process easier, please pre-register here: https://forms.gle/4Gc6SamK5thqB2Pk6 Please allow for a few minutes for the building security person to help you get to the 12th floor. Redhorse may email you a QR code to make check-in faster when you enter their office on 12th floor.

    5
  • 4th Meeting of Andrew Ng's "Neural Networks and Deep Learning"

    **** PRE-REGISTER HERE: https://forms.gle/xtj2r9Gk2BJJDkoE7 Please complete this simple form to speed up the check-in process. This will be our 4th meeting going through Andrew Ng's "Neural Networks and Deep Learning" online course (https://www.coursera.org/learn/neural-networks-deep-learning?specialization=deep-learning). We will meet again at Redhorse Corporation's Virginia offices, a short walk from the Rosslyn metro station. We are grateful to Redhorse Corporation for their help and generosity in providing us a place to meet so we can continue to learn and share ideas. ------------------------------------------------------------ About this meeting: We will first wrap up Week 3, "Shallow Neural Networks," (https://www.coursera.org/learn/neural-networks-deep-learning/home/week/3). Please watch the relevant videos (4 of them), starting with the week's 8th lecture, "Derivatives of activation functions". Next, one of our longtime members, George Zoto, will walk us through a Jupyter notebook that he has volunteered to put together to illustrate some of the concepts we have been learning in the course. Thanks, George! Finally, we will have an exciting special topic on reinforcement learning led by another of our members, Mark Shiffer. Mark plans to give us a brief introduction to reinforcement learning, and then present a game he chose to use for his exploration of reinforcement learning, and discuss the choices he made in setting up the problem and why. Thanks, Mark, for volunteering to do this! ------------------------------------------------------------ Some logistics: 1) The meeting will start at 6:00 pm, but feel free to arrive as early as 5:30 pm to mingle with colleagues. Please plan to come as early as possible, as it may be difficult to get entry to the Redhorse office after 6:00 pm. 2) Be sure to bring a photo ID. 3) Please pre-register here: https://forms.gle/xtj2r9Gk2BJJDkoE7 Completing this form enables Redhorse to email you a unique QR code. Please allow for a few minutes for the building security person to help you get to the 12th floor. Your QR code will make check-in faster when you enter Redhorse office.

    4
  • 3rd Meeting of Andrew Ng's "Neural Networks and Deep Learning"

    **** PRE-REGISTER HERE: https://forms.gle/1xbH5tGVL2YAQb9b8 Please complete this simple form to speed up the check-in process. This will be our 3rd meeting going through Andrew Ng's "Neural Networks and Deep Learning" online course (https://www.coursera.org/learn/neural-networks-deep-learning?specialization=deep-learning). We will meet again at Redhorse Corporation's Virginia offices, a short walk from the Rosslyn metro station. We thank Redhorse Corporation for their help and generosity in providing us a place to meet so we can continue to learn and share ideas. ------------------------------------------------------------ About this meeting: We will cover material from Week 3, "Shallow Neural Networks," (https://www.coursera.org/learn/neural-networks-deep-learning/home/week/3), starting with the week's 1st lecture, "Neural Network Overview," and progressing through the 5th, "Explanation for Vectorized Implementation." If there's time, we will cover the next 2 lectures on activation functions. Please watch these 7 videos before the meeting, and bring your ideas and questions about them. ------------------------------------------------------------ Some logistics: 1) The meeting will start at 5:50 pm, but feel free to arrive as early as 5:30 pm to mingle with colleagues. It may be difficult to get entry to the Redhorse office after 6:00 pm. 2) Be sure to bring a photo ID. 3) Please pre-register here: https://forms.gle/1xbH5tGVL2YAQb9b8 Completing this form enables Redhorse to email you a unique QR code. Please allow for a few minutes for the building security person to help you get to the 12th floor. Your QR code will make check-in faster when you enter Redhorse office.

    5
  • Special Topics: Applications of Deep Learning

    Redhorse Corporation

    Topic: Deep Learning in Healthcare **** PRE-REGISTER TO PROVIDE MORE INFO HERE: https://forms.gle/Cbv37B8fHoM2kvTv9 Please complete this simple form to speed up the check-in process. Wait listed attendees encourage to pre-register also. This will be the first of our Special Topics in Deep Learning which will include looking at applications of Deep Learning (DL) as well as hands-on projects using Deep Learning models. Our meetings will be discussion and helping each other get the models running. Please bring your laptop and be committed to reading the material and trying out the code or installs beforehand. Special Topic meetings will occur about every 6 weeks. Slack channels will be available for registered attendees for communicating in-between meetings. Agenda for this topic: * Discuss applications of DL in Healthcare * Review classic CNN architecture LeNet-5 * Get LeNet-5 running in Keras/Python on our laptops (or cloud). * Try training our CNN on labeled retina data. The first hour of meeting we will discuss how deep learning was applied in following healthcare applications: 1) Deep Learning for Detection of Diabetic Eye Disease Overview: https://optometry.berkeley.edu/big-data-takes-on-diabetic-retinopathy/ Approach: https://ai.googleblog.com/2016/11/deep-learning-for-detection-of-diabetic.html 2) Detecting Cancer Metastases on Gigapixel Pathology Images: https://arxiv.org/pdf/1703.02442.pdf 3) Scalable and Accurate Deep Learning with Electronic Health Records: https://www.nature.com/articles/s41746-018-0029-1.pdf The second hour we will discuss LeNet-5 CNN architecture and steps to install it on your laptop (or cloud) as time allows. 1) LeNet-5 [original 1998, paper by LeCun et al.] http://yann.lecun.com/exdb/publis/pdf/lecun-01a.pdf 2) Install Keras. 3) Install LeNet5 At the next special topics meetup after this one (6 weeks later), we will finish setup and installation and examine the retinal data for model training. A Slack channel will be used to communicate in-between. ------------------------------------------------------------ Our excellent host is Redhorse Corporation's Virginia offices, a short walk from the Rosslyn metro station. We thank Redhorse Corporation for their help and support in providing us a place to meet so we can continue to learn and share ideas. A couple of logistics items: 1) Feel free to arrive at 5:30 pm to allow for some time to get to the 12th floor and mingle with colleagues. We will begin the meeting at 6:00 pm. 2) Be sure to bring a photo ID. 3) To make the check-in process easier, please pre-register here: https://forms.gle/Cbv37B8fHoM2kvTv9 Completing this form will enable Redhorse to email you a unique QR code. Please allow for a few minutes for the building security person to help you get to the 12th floor. Your QR code will make check-in faster when you enter Redhorse office.

  • 2nd Meeting of Andrew Ng's course "Neural Networks and Deep Learning"

    **** PRE-REGISTER HERE: https://forms.gle/kGvgBBAzgTacwAXv7 Please complete this simple form to speed up the check-in process. This will be our 2nd meeting going through Andrew Ng's "Neural Networks and Deep Learning" online course (https://www.coursera.org/learn/neural-networks-deep-learning?specialization=deep-learning). As you may know, we have an excellent new venue for our meetups located at Redhorse Corporation's Virginia offices, a short walk from the Rosslyn metro station. We thank Redhorse Corporation for their help and generosity in providing us a place to meet so we can continue to learn and share ideas. ------------------------------------------------------------ About this meeting: In this meeting we will wrap up the Week 2 material with a focus on the lessons grouped under the heading "Python and Vectorization" (https://www.coursera.org/learn/neural-networks-deep-learning/lecture/NYnog/vectorization). Please watch the relevant videos (8 of them) ahead of time and come with your ideas and questions. ------------------------------------------------------------ A couple of logistics items: 1) Feel free to arrive at 5:30 pm to take some time to settle in and mingle with colleagues. We will begin the meeting at 6:00 pm. 2) Be sure to bring a photo ID. 3) To make the check-in process easier, please pre-register here: https://forms.gle/kGvgBBAzgTacwAXv7 Completing this form will enable Redhorse to create a unique QR code for you that they will e-mail to you. Bring that QR code with you when you arrive, and just scan it into their system to check in.

    10
  • Reboot Kickoff Meeting of DC Deep Learning Working Group

    Redhorse Corporation

    Calling All DC Deep Learning Enthusiasts! Winter is ending and we are coming out of hibernation and excited to announce that we are re-starting the DC Deep Learning Working Group meetings. We could not have done this without the generous help and sponsorship of Redhorse Corporation in Rosslyn. A big thank you to Redhorse for being our host for these meetings, and to Mark Shiffer for connecting us to them. Our plan for this series is to meet approximately every 2 weeks to discuss lectures in Andrew Ng's "Neural Networks and Deep Learning" online course (https://www.coursera.org/learn/neural-networks-deep-learning?specialization=deep-learning). In addition, every 3rd meeting we will devote to a special topic. This could be a hands-on exercise with data and code that we can tackle together, or a discussion of the technical details of one of the latest advances in deep learning applications. We are a "working group" so our meetings are not lectures, but rather group discussions of the meeting topic or exercise. Prior to each meeting, we will ask you to review the materials (lecture, code, papers) so that our discussion can be an interactive exchange of ideas and questions. RSVPs will open up 1 week in advance of our meeting. Please feel free to reach out to any of us with questions or suggestions or to say hello. We look forward to seeing you in a few weeks. Sincerely, Kathleen, Brian and Puneet Prerequisites: TBD

    9
  • DC Deep Learning Working Group Meetup at Walmart Labs

    This is a special co-hosted meeting of the DC Deep Learning Working Group and Walmart Labs. The DCDLWG is planning to resume our interactive, hands-on meetings again in August, but in the meanwhile, this is an opportunity to gather and hear about Walmart Lab's vision and approach to machine learning. Hope to see you there. Location: Walmart Labs, 10790 Parkridge Blvd, Suite 200, Reston, VA 20191 - Park in either side of the building - Walmart Labs Volunteers will receive and bring the associates for registration and to the meeting hall The agenda would be : 6:00pm - 7:00pm - Networking and light refreshments 7:00pm - 8:30pm - Presentation(s) 8:30pm - 9:00pm - Q & A Presentation 1: GPU Accelerated Computing for AI and beyond Presentation 2: The lifecycle of a data science project in Walmart, how machine learning platform is enabling data science teams to develop and deploy modes faster and how various teams are utilizing the platform. Speaker 1: Richard Ulrich Richard Ulrich has been at Walmart almost 28 years and has spent his entire career innovating in retail technology. Richard has innovated in each area of responsibility throughout his career and has approximately eighteen granted or pending patents in areas such general computing, RFID, analytics, and workforce management. Richard is currently working on high performance computing for advanced analytics, optimization, and machine learning, specifically focusing on the use of GPU to accelerate parallel computation. Speaker 2: Rakshith Uj Rakshith Uj is the product manager for the machine learning platform at Walmart Labs. He has been with Walmart for more than 4 years and has worked across customer experience, supply chain and machine learning as a product manager.

    3
  • 13th meeting of Designing, Visualizing & Understanding Deep Neural Networks

    Booz Allen Hamilton, DC Innovation Center

    This will be our 13th meeting going thru Berkeley's CS[masked] Designing, Visualizing and Understanding Deep Neural Networks (https://bcourses.berkeley.edu/courses/1453965/pages/cs294-129-designing-visualizing-and-understanding-deep-neural-networks). This meeting will be a discussion on Variational Bayesian Neural Networks lead by group member Patrick McLure from NIH. • Patrick's slides on Variational Bayesian Neural Networks (https://www.dropbox.com/sh/idlpyi6y2qzklff/AAC57MqCHJcdJDLiDFoh4ScAa?dl=0) Please try to review the slides prior to the meeting so that our discussion can be an interactive exchange of ideas and questions. ------------------------------------------------------------ We will continue to meet at the same location from the fall, the Booz Allen Hamilton Innovation Center near McPherson Square in DC. The DC Deep Learning Working Group kindly thanks Booz Allen Hamilton for their help and generosity in providing us a place to meet so we can continue to learn and share ideas. Please send us your full name before the meeting, if it is not the same as your Meetup name, and also bring a photo ID to the meeting for entrance to the facility. BAH employees will meet you in the lobby and escort you up to the meeting room. Also, please change your RSVP if you find you cannot make the meeting so someone from wait-list can take your spot. Thanks also to Deep Learning Analytics! --------------------------------------------------------------------------------------- Deep Networks have revolutionized computer vision, speech recognition and language translation. They have growing impact in many areas of science and engineering. They also do not follow a closed set of theoretical principles. In Yann Lecun's words they require "an interplay between intuitive insights, theoretical modeling, practical implementations, empirical studies, and scientific analyses." This course attempts to cover that ground, and has three goals: * Design principles and best practices: design motifs that work well in particular domains, structure optimization and parameter optimization. * Visualizing deep networks. Exploring the training and use of deep networks with visualization tools. * Understanding deep networks. Methods with formal guarantees: generative and adversarial models, tensor factorization. This course draws heavily from the excellent Stanford course CS231n: Convolutional Neural Networks for Visual Recognition (Links to an external site.) (http://cs231n.stanford.edu/) by Li, Karpathy and Johnson and on the excellent online book, Deep Learning (http://www.deeplearningbook.org/), by Ian Goodfellow, Yoshua Bengio, and Aaron Courville ------------------------------------------------------------ GitHub: We will continue to add content to our shared GitHub repository (https://github.com/dcdlwg/stanford_cs224d) at DC DLWG (https://github.com/dcdlwg). The solutions to any problems we go over in our meetings will be uploaded to the repo, and we will continue to share resources and code there as we progress through the course. You can share your own work by creating your own repo at DC DLWG (https://github.com/dcdlwg). So please create a GitHub account when you get a chance, if you don’t have one already. Prerequisites: We try to accommodate all levels of experience but the course assumes familiarity with the following: Proficiency in Python, calculus, linear algebra, basic probability and statistics and some concepts of machine learning. Please look through the course website (https://bcourses.berkeley.edu/courses/1453965) to see if this is the right level for you.

  • 12th meeting of Designing, Visualizing & Understanding Deep Neural Networks

    This will be our 12th meeting going thru Berkeley's CS[masked] Designing, Visualizing and Understanding Deep Neural Networks (https://bcourses.berkeley.edu/courses/1453965/pages/cs294-129-designing-visualizing-and-understanding-deep-neural-networks). 11/9 Lecture, "Variational Autoencoders" ------------------------------------------------------------ We will continue to meet at the same location from the fall, the Booz Allen Hamilton Innovation Center near McPherson Square in DC. The DC Deep Learning Working Group kindly thanks Booz Allen Hamilton for their help and generosity in providing us a place to meet so we can continue to learn and share ideas. Please send us your full name before the meeting, if it is not the same as your Meetup name, and also bring a photo ID to the meeting for entrance to the facility. BAH employees will meet you in the lobby and escort you up to the meeting room. Also, please change your RSVP if you find you cannot make the meeting so someone from wait-list can take your spot. Thanks also to Deep Learning Analytics! ------------------------------------------------------------ About this meeting: For our 12th meeting we will discuss the 11/9 Lecture on "Variational Autoencoders". (given by Durk Kingma.) LOOK UPDATE: Instead use Stanford CS231n lecture 13. Video is (https://youtu.be/5WoItGTWV54) • Video for Variational Autoencoders. (https://youtu.be/prCVH1_3AqQ?t=8m10s) • PDF slides of Variational Autoencoders lecture. (https://bcourses.berkeley.edu/courses/1453965/files/70020222/download?wrap=1) Please review the lecture prior to the meeting so that our discussion can be an interactive exchange of ideas and questions. For those jumping in for the first time we also ask that you familiarize yourself with material in previous lectures so you are up to speed. All the course lectures videos, (starting on lecture 4) are archived here (https://www.youtube.com/playlist?list=PLkFD6_40KJIxopmdJF_CLNqG3QuDFHQUm). Homepage for the Berkeley course with additional material here. (https://bcourses.berkeley.edu/courses/1453965/pages/cs294-129-designing-visualizing-and-understanding-deep-neural-networks) Deep Networks have revolutionized computer vision, speech recognition and language translation. They have growing impact in many areas of science and engineering. They also do not follow a closed set of theoretical principles. In Yann Lecun's words they require "an interplay between intuitive insights, theoretical modeling, practical implementations, empirical studies, and scientific analyses." This course attempts to cover that ground, and has three goals: * Design principles and best practices: design motifs that work well in particular domains, structure optimization and parameter optimization. * Visualizing deep networks. Exploring the training and use of deep networks with visualization tools. * Understanding deep networks. Methods with formal guarantees: generative and adversarial models, tensor factorization. This course draws heavily from the excellent Stanford course CS231n: Convolutional Neural Networks for Visual Recognition (Links to an external site.) (http://cs231n.stanford.edu/) by Li, Karpathy and Johnson and on the excellent online book, Deep Learning (http://www.deeplearningbook.org/), by Ian Goodfellow, Yoshua Bengio, and Aaron Courville ------------------------------------------------------------ GitHub: We will continue to add content to our shared GitHub repository (https://github.com/dcdlwg/stanford_cs224d) at DC DLWG (https://github.com/dcdlwg). The solutions to any problems we go over in our meetings will be uploaded to the repo, and we will continue to share resources and code there as we progress through the course. You can share your own work by creating your own repo at DC DLWG (https://github.com/dcdlwg). So please create a GitHub account when you get a chance, if you don’t have one already. Prerequisites: We try to accommodate all levels of experience but the course assumes familiarity with the following: Proficiency in Python, calculus, linear algebra, basic probability and statistics and some concepts of machine learning. Please look through the course website (https://bcourses.berkeley.edu/courses/1453965) to see if this is the right level for you.

    4
  • 11th meeting of Designing, Visualizing & Understanding Deep Neural Networks

    This will be our 11th meeting going thru Berkeley's CS[masked] Designing, Visualizing and Understanding Deep Neural Networks (https://bcourses.berkeley.edu/courses/1453965/pages/cs294-129-designing-visualizing-and-understanding-deep-neural-networks). Lecture 16, "Deep Reinforcement Learning part 2" ------------------------------------------------------------ HEADS UP! We are back to our original location, the Booz Allen Hamilton Innovation Center near McPherson Square in DC. The DC Deep Learning Working Group kindly thanks Booz Allen Hamilton for their help and generosity in providing us a place to meet so we can continue to learn and share ideas. Please send us your full name before the meeting, if it is not the same as your Meetup name, and also bring a photo ID to the meeting for entrance to the facility. BAH employees will meet you in the lobby and escort you up to the meeting room. Also, please change your RSVP if you find you cannot make the meeting so someone from wait-list can take your spot. Thanks also to Deep Learning Analytics! ------------------------------------------------------------ About this meeting: For our 11th meeting we will discuss Lecture 16, "Deep Reinforcement Learning II" . (Note this is the one given by John Schulman.) • Video for Lecture 16, Deep Reinforcement Learning II ( https://www.youtube.com/watch?v=aRNtyRQKE7Q&list=PLkFD6_40KJIxopmdJF_CLNqG3QuDFHQUm&index=14 ) (starts around 20 min in) • Slides for Lecture 16, Deep Reinforcement Learning II (https://bcourses.berkeley.edu/courses/1453965/files/69929026/download?verifier=Ikf9yI9wRqaNTMu2zqvtV0S8McteAeYfZybfBzCm&wrap=1) Please review the lecture prior to the meeting so that our discussion can be an interactive exchange of ideas and questions. For those jumping in for the first time we also ask that you familiarize yourself with material in previous lectures so you are up to speed. All the course lectures videos, (starting on lecture 4) are archived here (https://www.youtube.com/playlist?list=PLkFD6_40KJIxopmdJF_CLNqG3QuDFHQUm). Homepage for the Berkeley course with additional material here. (https://bcourses.berkeley.edu/courses/1453965/pages/cs294-129-designing-visualizing-and-understanding-deep-neural-networks) Deep Networks have revolutionized computer vision, speech recognition and language translation. They have growing impact in many areas of science and engineering. They also do not follow a closed set of theoretical principles. In Yann Lecun's words they require "an interplay between intuitive insights, theoretical modeling, practical implementations, empirical studies, and scientific analyses." This course attempts to cover that ground, and has three goals: * Design principles and best practices: design motifs that work well in particular domains, structure optimization and parameter optimization. * Visualizing deep networks. Exploring the training and use of deep networks with visualization tools. * Understanding deep networks. Methods with formal guarantees: generative and adversarial models, tensor factorization. This course draws heavily from the excellent Stanford course CS231n: Convolutional Neural Networks for Visual Recognition (Links to an external site.) (http://cs231n.stanford.edu/) by Li, Karpathy and Johnson and on the excellent online book, Deep Learning (http://www.deeplearningbook.org/), by Ian Goodfellow, Yoshua Bengio, and Aaron Courville ------------------------------------------------------------ GitHub: We will continue to add content to our shared GitHub repository (https://github.com/dcdlwg/stanford_cs224d) at DC DLWG (https://github.com/dcdlwg). The solutions to any problems we go over in our meetings will be uploaded to the repo, and we will continue to share resources and code there as we progress through the course. You can share your own work by creating your own repo at DC DLWG (https://github.com/dcdlwg). So please create a GitHub account when you get a chance, if you don’t have one already. Prerequisites: We try to accommodate all levels of experience but the course assumes familiarity with the following: Proficiency in Python, calculus, linear algebra, basic probability and statistics and some concepts of machine learning. Please look through the course website (https://bcourses.berkeley.edu/courses/1453965) to see if this is the right level for you.

    4