MIT's introductory course on deep learning methods with applications to computer vision, natural language processing, biology, and more! Students will gain foundational knowledge of deep learning algorithms and get practical experience in building neural networks in TensorFlow. Course concludes with a project proposal competition with feedback from staff and panel of industry sponsors. Prerequisites assume calculus (i.e. taking derivatives) and linear algebra (i.e. matrix multiplication), we'll try to explain everything else along the way! Experience in Python is helpful but not necessary. Listeners are welcome!
1:00pm-4:00pm, MIT Room 32-123
1:00pm-1:45pm: Lecture Part 1
1:45pm-2:30pm: Lecture Part 2
2:30pm-2:40pm: Snack Break
2:40pm-4:00pm: Software Labs
We are expecting very elementary knowledge of linear algebra and calculus. How to multiply matrices, take derivatives and apply the chain rule. Familiarity in Python is a big plus as well. The course will be beginner friendly since we have many registered students from outside of computer science.
If you would like to receive course related updates and lecture materials please sign up for our mailing list.
All course materials available online for free but are copyrighted and licensed under the MIT license. If you are an instructor and would like to use any materials from this course (slides, labs, code), you must add the following reference to each slide:
© Alexander Amini and Ava Soleimany
MIT 6.S191: Introduction to Deep Learning
IntroToDeepLearning.com
All course materials are copyrighted and licensed under the MIT license. If you are an instructor and would like to use any materials from this course (slides, labs, code), you must add the following reference to each slide:
© Alexander Amini and Ava Soleimany
MIT 6.S191: Introduction to Deep Learning
IntroToDeepLearning.com
If you are an MIT student, postdoc, faculty, or affiliate and would like to become involved with this course please email introtodeeplearning-staff@mit.edu. We are always accepting new applications to join the course staff.
This class would not be possible without our amazing sponsors and has been sponsored by Google, IBM, NVIDIA, and Onepanel. If you are interesting in becoming involved in this course as a sponsor please contact us at introtodeeplearning-staff@mit.edu .
Riyadh Baghdadi
Blake Elias
Kristian Georgiev
Shinjini Ghosh
Hunter Hansen
Konstantin Krismer
Alana Marzoev
Julia Moseyko
Jacob Phillips
Monisha Pushpanathan
Roshni Sahoo
Andy Shea
Gilbert Yang
Copyright © MIT 6.S191. banner image; page template
Artificial intelligence and machine learning have experienced a renaissance in the past decade, thanks largely to the success of deep learning methods. However, while deep learning has proven itself to be extremely powerful, most of today’s most successful deep learning systems suffer from a number of important limitations, ranging from the requirement for enormous training data sets to lack of interpretability to vulnerability to "hacking" via adversarial examples. In my talk, I will survey some of these limitations and propose that one path forward involves building hybrid systems that combine neural networks with techniques and ideas from symbolic AI, a parallel tradition of AI whose origins date back to the beginning of AI. I will show example neurosymbolic hybrid systems where neural networks and symbolic systems complement each other’s strengths and weaknesses, enabling systems that are accurate, sample efficient, and interpretable. Finally, I will show other directions we are pursuing in the space of neural-symbolic hybrid systems, and argue that these methods at the intersection provide a powerful path forward for the broad adoption of AI.
David Cox is the IBM Director of the MIT-IBM Watson AI Lab, a first of its kind industry-academic collaboration between IBM and MIT, focused on fundamental research in artificial intelligence. David's ongoing research is primarily focused on bringing insights from neuroscience into machine learning and computer vision research. His work has spanned a variety of disciplines, from imaging and electrophysiology experiments in living brains, to the development of machine learning and computer vision methods, to applied machine learning and high performance computing methods.
Data-driven methods in Robotics circumvent hand-tuned feature engineering, albeit lack guarantees and often incur a massive computational expense. My research aims to bridge this gap and enable generalizable imitation for robot autonomy. We need to build systems that can capture semantic task structures that promote sample efficiency and can generalize to new task instances across visual, dynamical or semantic variations. And this involves designing algorithms that unify learning with perception, control and planning. In this talk, I will how inductive biases and priors help with Generalizable Autonomy. First I will talk about choice of action representations in RL and imitation from ensembles of suboptimal supervisors. Then I will talk about latent variable models in self-supervised learning. Finally I will talk about meta-learning for multi-task learning and data gather in robotics.
Animesh Garg is a CIFAR AI Chair Assistant Professor of at University of Toronto and Vector Institute. He is also a Senior Research Scientist at Nvidia. His research interests focus on intersection of Learning & Perception in Robot Manipulation. He works on efficient generalization in large scale imitation learning. Animesh works applications of robot manipulation in surgery and manufacturing as well as personal robotics. Previously, Animesh received his Ph.D. from the University of California, Berkeley and a postdoc at Stanford AI Labs. His work has won multiple best paper awards and nominations including ICRA 2019, ICRA 2015 and IROS 2019, among others and has also featured in press outlets such as New York Times, BBC, and Wired.
In this talk, we will review modern rendering techniques and discuss how deep learning can extend the gamut of this long-lasting research topic. We will investigate deep neural networks as 1) plug-and-play sub-modules that reduce the cost of physically-based rendering; 2) end-to-end pipelines that inspire novel graphics applications. In particular, we will focus on "differentiable rendering," a methodology that solves complex inverse graphics problems and achieved great success in scene reconstruction, generation, and depiction.
Chuan Li is a research scientist at Lambda Labs. His work focuses specifically on the convergent field of computer graphics, computer vision, and machine learning. He completed his Ph.D. in image-based modeling at the University of Bath. Before joining Lambda Labs, he was a Postdoc researcher at Max Planck Institute of Informatics and a research associate at Utrecht University and Mainz University. His research in visual data analysis and synthesis was published at CVPR, ICCV, ECCV, NIPS, Siggraph.
Predicting the relationship between a molecule's structure and its odor remains a difficult, decades-old task. This problem, termed quantitative structure-odor relationship (QSOR) modeling, is an important challenge in chemistry, impacting human nutrition, manufacture of synthetic fragrance, the environment, and sensory neuroscience. We use of graph neural networks for QSOR, and show they significantly out-perform prior methods on a novel data set labeled by olfactory experts. Additional analysis shows that the learned embeddings from graph neural networks capture a meaningful odor space representation of the underlying relationship between structure and odor, as demonstrated by strong performance on two challenging transfer learning tasks. Machine learning has already had a large impact on the senses of sight and sound. Based on these early results with graph neural networks for molecular properties, we hope machine learning can eventually do for olfaction what it has already done for vision and hearing.
Alex Wiltschko is a research scientist at Google Brain, focusing on building more flexible machine learning software systems, and also applications of machine learning to biology. He has helped build several machine learning libraries, including torch-autograd, and Tangent, a compiler-based autodiff library for Python at Google. He completed his PhD in Neurobiology at Harvard, focusing on quantifying behavior and body language using depth cameras and nonparametric time-series modeling.