Neural Networks and Deep Learning
Columbia University course ECBM E4040
Course in a nutshell:
Theoretical underpinnings and practical aspects of Neural Networks and Deep Learning. Convolutional and Recurrent Neural Networks. Focus on applications and projects.
Bulletin Description:
 Developing features & internal representations of the world, artificial neural networks, classifying handwritten digits with logistics regression, feedforward deep networks, back propagation in multilayer perceptrons, regularization of deep or distributed models, optimization for training deep models, convolutional neural networks, recurrent and recursive neural networks, deep learning in speech and object recognition.

Course:
 Fall 2019: Fridays 10:1012:20
 Fall 2018: Fridays 10:1012:20 (1) Course webpage, (2) Syllabus
 Fall 2017: Fridays 10:1012:20.
 Fall 2016: Fridays 10:1012:20.
 Required prerequisites: knowledge of linear algebra, probability and statistics, programming. Strongly recommended: machine learning.
 Possible prerequisite courses: (BMEB W4020) or (BMEE E4030) or (ECBM E4090) or (EECS E4750) or (COMS W4771) or similar.

Content
 Introduction to neural networks.
 Convolutional and recurrent neural networks.
 Focuses on the intuitive understanding of deep learning.
 Review of underpinning theory  linear algebra, statistics, machine learning.
 Analytical study and software design.
 Threefour assignments in Python and one DL framework (Tensorflow or PyTorch)
 Significant project.
 Enables further exploration of key concepts in deep learning.

Organization
 Lectures:
 Presentation of material by instructors and guest lecturers
 Homeworks:
 Combination of analytical and programming assignments
 Projects:
 Teambased
 Students with complementary backgrounds
 Significant design
 Reports and presentations to Columbia and NYC community
 Best could qualify for publications and/or funding
 Industry participation:
 Project definition and sponsoring
 Weekly presentations
 Interaction with students through mentoring

Project Areas
 Medical
 Autonomous cars
 Environmental
 Smart cities
 Physical data analytics

Books, Tools and Resources
 BOOKS:
 2018, 2017 software platform:
 Google TensorFlow, Google Cloud, Python, bitbucket
 2016 software platform:

2019 Projects

2018 Projects
 A deep learning framework for relationship extraction from articles using longshort term memory and named entity recognition
 A Neural Algorithm of Artistic Style
 A Neural Representation of Sketch Drawings
 Adversarial Variational Bayes: Unifying Variational Autoencoders and Generative Adversarial Networks
 Backprop KF: Learning Discriminative Deterministic State Estimators
 Deep contextualized word representations
 Dynamic Routing Between Capsules
 Gesture Recognition
 Learned in Translation: Contextualized Word Vectors
 Learning a Probabilistic Latent Space of Object Shapes via 3D GenerativeAdversarial Modeling
 Maximum Classifier Discrepancy for Unsupervised Domain Adaptation
 MultiDigit Number Recognition from Street View Imagery Using Deep Convolutional Neural Networks
 Neural Networks for Automated Essay Grading
 Parallel MultiDimensional LSTM,With Application to Fast Biomedical Volumetric Image Segmentation
 PixelGAN Autoencoders
 Prevention of catastrophic forgetting in Neural Networks for lifelong learning
 Semantic Image Inpainting with Deep Generative Models
 Towards Accurate Binary Convolutional Neural Network
 Universal Style Transfer via Feature Transforms
 Unsupervised ImagetoImage Translation Networks

2016 Projects
 Striving for Simplicity: The All Convolutional Net
 A Combined Semisupervised Learning mechanism for Video Data via Deep Learning
 A Neural Algorithm of Artistic Style
 Adieu features? Endtoend speech emotion recognition using a deep convolutional recurrent network
 Colorful Image Colorization
 Deep Networks with Stochastic Depth
 Highway Networks
 Image SuperResolution Using Deep Convolutional Networks
 Learning to Protect Communications with Adversarial Neural Cryptography
 Singing Voice Separation from Monaural Recordings Using Deep
 Recurrent Neural Networks
 Spatial Transformer Networks
 Spoken Language Understanding Using LongShort Term Memory Neural Networks
 Striving for Simplicity: The All Convolutional Net
 Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks

2017 Projects
 A Neural Algorithm of Artistic Style
 BinaryConnect: Training Deep Neural Networks with Binary Weights during Propagations
 Composing music with recurrent neural networks
 Deep Learning Face Representation from Predicting 10,000 Classes
 Deep Learning in Finance; Deep Portfolio Theory
 Deep Networks with Stochastic Depth
 DeepDriving: Deep Learning for Autonomous Driving
 Depth Map Prediction from a Single Image using a MultiScale Deep Network
 Draw: A Recurrent Neural Network for Image Generation
 Long Short Term Memory Networks for Anomaly Detection in Time Series
 Multidigit Number Recognition from Street View Imagery using Deep Convolutional Neural Networks
 Predicting HIV Risk Factors From Unstructured Clinical Text
 Richer Convolutional Features for Edge Detection
 Spectral Representations for Convolutional Neural Networks
 Understanding Deep Learning Requires Rethinking Generalization
 Using Convolutional Networks and Satellite Imagery to Identify Patterns in Urban Environments at a Large Scale

Course sponsored by equipment and financial contributions of:
 NVidia GPU Education Center, Google Cloud, IBM Bluemix, AWS Educate, Atmel, Broadcom (Wiced platform); Intel (Edison IoT platform), Silicon Labs.
