MIT 6.S191 (2020): Introduction to Deep Learning Alexander Amini 2020-02-08 | MIT Introduction to Deep Learning 6.S191: Lecture 1Foundations of Deep LearningLecturer: Alexander AminiJanuary 2020For all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline0:00 - Introduction4:14 - Course information8:10 - Why deep learning?11:01 - The perceptron13:07 - Activation functions15:32 - Perceptron example18:54 - From perceptrons to neural networks25:23 - Applying neural networks28:16 - Loss functions31:14 - Training and gradient descent35:13 - Backpropagation39:25 - Setting the learning rate43:43 - Batched gradient descent46:46 - Regularization: dropout and early stopping51:58 - SummarySubscribe to @stay up to date with new deep learning lectures at MIT, or follow us on @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191: The Future of Robot Learning Alexander Amini 2023-05-12 | MIT Introduction to Deep Learning 6.S191: Lecture 10The Future of Robot LearningLecturer: Daniela Rus2023 EditionFor all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline - coming soon!Subscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191: The Modern Era of Statistics Alexander Amini 2023-05-05 | MIT Introduction to Deep Learning 6.S191: Lecture 9The Modern Era of StatisticsLecturer: Ramin Hasani2023 EditionFor all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline - coming soon!Subscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191: Text-to-Image Generation Alexander Amini 2023-04-28 | MIT Introduction to Deep Learning 6.S191: Lecture 8Deep Learning Limitations and New FrontiersLecturer: Dilip Krishnan2023 EditionFor all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline - coming soon!Subscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191: Deep Learning New Frontiers Alexander Amini 2023-04-22 | MIT Introduction to Deep Learning 6.S191: Lecture 7Deep Learning Limitations and New FrontiersLecturer: Ava Amini2023 EditionFor all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline - coming soon!Subscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191: Reinforcement Learning Alexander Amini 2023-04-14 | MIT Introduction to Deep Learning 6.S191: Lecture 5Deep Reinforcement LearningLecturer: Alexander Amini2023 EditionFor all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline:0:00 - Introduction 3:49 - Classes of learning problems6:48 - Definitions12:24 - The Q function17:06 - Deeper into the Q function21:32 - Deep Q Networks29:15 - Atari results and limitations32:42 - Policy learning algorithms36:42 - Discrete vs continuous actions39:48 - Training policy gradients47:17 - RL in real life49:55 - VISTA simulator52:04 - AlphaGo and AlphaZero and MuZero56:34 - SummarySubscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191: Robust and Trustworthy Deep Learning Alexander Amini 2023-04-07 | MIT Introduction to Deep Learning 6.S191: Lecture 5Robust and Trustworthy Deep LearningLecturer: Sadhana Lolla (Themis AI, themisai.io)2023 EditionFor all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline0:00 - Introduction and Themis AI3:46 - Background7:29 - Challenges for Robust Deep Learning8:24 - What is Algorithmic Bias? 14:13 - Class imbalance16:25 - Latent feature imbalance20:30 - Debiasing variational autoencoder (DB-VAE)23:24 - DB-VAE mathematics27:40 - Uncertainty in deep learning29:50 - Types of uncertainty in AI32:48 - Aleatoric vs epistemic uncertainty33:29 - Estimating aleatoric uncertainty37:42 - Estimating epistemic uncertainty44:11 - Evidential deep learning46:44 - Recap of challenges47:14 - How Themis AI is transforming risk-awareness of AI49:30 - Capsa: Open-source risk-aware AI wrapper51:51 - Unlocking the future of trustworthy AISubscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191: Deep Generative Modeling Alexander Amini 2023-03-31 | MIT Introduction to Deep Learning 6.S191: Lecture 4Deep Generative ModelingLecturer: Ava Amini2023 EditionFor all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline0:00 - Introduction5:48 - Why care about generative models? 7:33 - Latent variable models9:30 - Autoencoders15:03 - Variational autoencoders21:45 - Priors on the latent distribution28:16 - Reparameterization trick31:05 - Latent perturbation and disentanglement 36:37 - Debiasing with VAEs38:55 - Generative adversarial networks41:25 - Intuitions behind GANs44:25 - Training GANs50:07 - GANs: Recent advances50:55 - Conditioning GANs on a specific label53:02 - CycleGAN of unpaired translation56:39 - Summary of VAEs and GANs57:17 - Diffusion Model sneak peakSubscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191: Convolutional Neural Networks Alexander Amini 2023-03-24 | MIT Introduction to Deep Learning 6.S191: Lecture 3Convolutional Neural Networks for Computer VisionLecturer: Alexander Amini2023 EditionFor all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline0:00 - Introduction2:37 - Amazing applications of vision5:35 - What computers "see"12:38- Learning visual features17:51 - Feature extraction and convolution22:23 - The convolution operation27:30 - Convolution neural networks34:29 - Non-linearity and pooling40:07 - End-to-end code example41:23 - Applications43:18 - Object detection51:36 - End-to-end self driving cars54:08 - SummarySubscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191: Recurrent Neural Networks, Transformers, and Attention Alexander Amini 2023-03-17 | MIT Introduction to Deep Learning 6.S191: Lecture 2Recurrent Neural NetworksLecturer: Ava Amini 2023 EditionFor all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline0:00 - Introduction3:07 - Sequence modeling5:09 - Neurons with recurrence12:05 - Recurrent neural networks13:47 - RNN intuition15:03 - Unfolding RNNs18:57 - RNNs from scratch21:50 - Design criteria for sequential modeling23:45 - Word prediction example29:57 - Backpropagation through time32:25 - Gradient issues37:03 - Long short term memory (LSTM)39:50 - RNN applications44:50 - Attention fundamentals 48:10 - Intuition of attention50:30 - Attention and search relationship52:40 - Learning attention with neural networks58:16 - Scaling attention and applications1:02:02 - SummarySubscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT Introduction to Deep Learning | 6.S191 Alexander Amini 2023-03-10 | MIT Introduction to Deep Learning 6.S191: Lecture 1*New 2023 Edition*Foundations of Deep LearningLecturer: Alexander AminiFor all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline0:00 - Introduction8:14 - Course information11:33 - Why deep learning?14:48 - The perceptron20:06 - Perceptron example23:14 - From perceptrons to neural networks29:34 - Applying neural networks 32:29 - Loss functions35:12 - Training and gradient descent40:25 - Backpropagation44:05 - Setting the learning rate48:09 - Batched gradient descent51:25 - Regularization: dropout and early stopping57:16 - SummarySubscribe to stay up to date with new deep learning lectures at MIT, or follow us on @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191: Uncertainty in Deep Learning Alexander Amini 2022-05-28 | MIT Introduction to Deep Learning 6.S191: Lecture 10Uncertainty in Deep LearningLecturer: Jasper Snoek (Research Scientist, Google Brain)Google BrainJanuary 2022For all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline - coming soon!Subscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191: AI for Science Alexander Amini 2022-05-13 | MIT Introduction to Deep Learning 6.S191: Lecture 9AI for ScienceLecturer: Anima Anandkumar (Director of ML Research, NVIDIA)NVIDIA ResearchJanuary 2022For all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline - coming soon!Subscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191: Automatic Speech Recognition Alexander Amini 2022-05-02 | MIT Introduction to Deep Learning 6.S191: Lecture 8How Rev.com harnesses human-in-the-loop and deep learning to build the world's best English speech recognition engineLecturer: Miguel Jetté (Head of AI R&D) and Jennifer Drexler (Senior Speech Scientist)Rev.AIJanuary 2022For all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline - coming soon!Subscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191: LiDAR for Autonomous Driving Alexander Amini 2022-04-22 | MIT Introduction to Deep Learning 6.S191: Lecture 7Deep Learning for Autonomous DrivingLecturer: Omer Keilaf (CEO) and Amir Day (Head of CV & DL)Innoviz TechnologiesJanuary 2022For all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline - coming soon!Subscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191 (2022): Deep Learning New Frontiers Alexander Amini 2022-04-15 | MIT Introduction to Deep Learning 6.S191: Lecture 6Deep Learning Limitations and New FrontiersLecturer: Ava SoleimanyJanuary 2022For all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline - coming soon!Subscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191 (2022): Reinforcement Learning Alexander Amini 2022-04-08 | MIT Introduction to Deep Learning 6.S191: Lecture 5Deep Reinforcement LearningLecturer: Alexander AminiJanuary 2022For all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline - coming soon!Subscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191 (2022): Deep Generative Modeling Alexander Amini 2022-04-01 | MIT Introduction to Deep Learning 6.S191: Lecture 4Deep Generative ModelingLecturer: Ava SoleimanyJanuary 2022For all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline - coming soon!Subscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191 (2022): Convolutional Neural Networks Alexander Amini 2022-03-25 | MIT Introduction to Deep Learning 6.S191: Lecture 3Convolutional Neural Networks for Computer VisionLecturer: Alexander AminiJanuary 2022For all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline - coming soon!Subscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191 (2022): Recurrent Neural Networks and Transformers Alexander Amini 2022-03-18 | MIT Introduction to Deep Learning 6.S191: Lecture 2Recurrent Neural NetworksLecturer: Ava SoleimanyJanuary 2022For all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline0:00 - Introduction1:59 - Sequence modeling4:16 - Neurons with recurrence10:09 - Recurrent neural networks11:42 - RNN intuition14:44 - Unfolding RNNs16:43 - RNNs from scratch19:49 - Design criteria for sequential modeling21:00 - Word prediction example27:49 - Backpropagation through time30:02 - Gradient issues33:53 - Long short term memory (LSTM)35:35 - RNN applications40:22 - Attention fundamentals 43:12 - Intuition of attention44:53 - Attention and search relationship47:16 - Learning attention with neural networks54:52 - Scaling attention and applications56:09 - SummarySubscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT Introduction to Deep Learning (2022) | 6.S191 Alexander Amini 2022-03-11 | MIT Introduction to Deep Learning 6.S191: Lecture 1Foundations of Deep LearningLecturer: Alexander AminiFor all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline0:00 - Introduction6:35 - Course information9:51 - Why deep learning?12:30 - The perceptron14:31 - Activation functions17:03 - Perceptron example20:25 - From perceptrons to neural networks26:37 - Applying neural networks 29:18 - Loss functions31:19 - Training and gradient descent35:46 - Backpropagation38:55 - Setting the learning rate41:37 - Batched gradient descent43:45 - Regularization: dropout and early stopping47:58 - SummarySubscribe to stay up to date with new deep learning lectures at MIT, or follow us on @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
LiDAR-Based End-to-End Navigation | ICRA 2021 Alexander Amini 2021-05-23 | This video is part of the paper: "Efficient and Robust LiDAR-Based End-to-End Navigation" which is presented at the International Conference on Robotics and Automation (ICRA) 2021. Efficient and Robust LiDAR-Based End-to-End NavigationZhijian Liu^, Alexander Amini^, Sibo Zhu, Sertac Karaman, Song Han, and Daniela L. RusPaper Link: arxiv.org/abs/2105.09932Website: https://le2ed.mit.eduAbstract: Deep learning has been used to demonstrate end-to-end neural network learning for autonomous vehicle control from raw sensory input. While LiDAR sensors provide reliably accurate information, existing end-to-end driving solutions are mainly based on cameras since processing 3D data requires a large memory footprint and computation cost. On the other hand, increasing the robustness of these systems is also critical; however, even estimating the model’s uncertainty is very challenging due to the cost of sampling-based methods. In this paper, we present an efficient and robust LiDAR-based end-to-end navigation framework. We first introduce Fast-LiDARNet that is based on sparse convolution kernel optimization and hardware-aware model design. We then propose Hybrid Evidential Fusion that directly estimates the uncertainty of the prediction from only a single forward pass and then fuses the control predictions intelligently. We evaluate our system on a full-scale vehicle and demonstrate lane-stable as well as navigation capabilities. In the presence of out-of-distribution events (e.g., sensor failures), our system significantly improves robustness and reduces the number of takeovers in the real world.
MIT 6.S191: AI in Healthcare Alexander Amini 2021-04-23 | MIT Introduction to Deep Learning 6.S191: Lecture 12AI in HealthcareLecturer: Dr. Katherine Chou, Google BrainJanuary 2021For all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline0:00 - Introduction2:40 - Applications of AI in healthcare5:12 - End-to-end lung cancer screening6:05 - Pathology8:02 - Genomics11:44 - Higher quality and more equitable learning13:34 - Moonshots at Google16:43 - Generating labels, bias, and uncertainty20:22 - Plan for model limitations23:31 - Healthcare patient vs person27:35 - Summary and conclusion28:03 - Poll results and Q&ASubscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191: Towards AI for 3D Content Creation Alexander Amini 2021-04-16 | MIT Introduction to Deep Learning 6.S191: Lecture 11Towards AI for 3D Content CreationLecturer: Prof. Sanja Fidler, University of Toronto and NVIDIAJanuary 2021For all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline0:00 - Introduction2:10 - What is 3D content? 7:00 - AI for 3D content creation8:20 - Synthesizing worlds11:45 - Scene composition15:50 - Learning structure20:55 - Synthesizing medical data23:00 - Recovering rules of the world26:45 - Object creation28:50 - Graphics via differentiable rendering32:30 - Data generation38:30 - Neural simulation43:25 - 3D deep learning library45:25 - Summary and conclusionSubscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191: Taming Dataset Bias via Domain Adaptation Alexander Amini 2021-04-09 | MIT Introduction to Deep Learning 6.S191: Lecture 10Taming Dataset Bias via Domain AdaptationLecturer: Prof. Kate Saenko, MIT-IBM Watson AI LabJanuary 2021For all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline0:00 - Introduction3:20 - When does dataset bias occur? 7:00 - Implications in the real-world12:41 - Dealing with data bias14:38 - Adversarial domain alignment20:30 - Pixel space alignment26:03 - Few-shot pixel alignment33:56 - Moving beyond alignment38:59 - Enforcing consistency42:05 - Summary and conclusionSubscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191: Deep CPCFG for Information Extraction Alexander Amini 2021-04-02 | MIT Introduction to Deep Learning 6.S191: Lecture 9Deep CPCFG for Information ExtractionLecturer: Nigel Duffy and Freddy Chua, Ernst & Young AI LabsJanuary 2021For all lectures, slides, and lab materials: http://introtodeeplearning.comMore details on Deep Conditional Probabilistic Context Free Grammars (CPCFG): arxiv.org/abs/2103.05908Code and datasets: github.com/deepcpcfg/datasetsLecture Outline0:00 - Introduction4:18 - What is information extraction?7:19 - Types of information (headers, line items, etc)11:57 - Representing document schemas12:35 - Philosophy of end-to-end deep learning16:38 - Context free grammars (CFG)20:55 - Parsing with deep learning27:10 - Learning objective and training28:21 - 2 dimensional parsing33:20 - Handling noise in the parsing35:23 - Experimental results38:00 - Question and answeringSubscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191: AI Bias and Fairness Alexander Amini 2021-03-26 | MIT Introduction to Deep Learning 6.S191: Lecture 8Algorithmic Bias and FairnessLecturer: Ava SoleimanyJanuary 2021For all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline0:00 - Introduction and motivation1:40 - What does "bias" mean?4:22 - Bias in machine learning8:32 - Bias at all stages in the AI life cycle9:25 - Outline of the lecture10:00 - Taxonomy (types) of common biases11:29 - Interpretation driven biases16:04 - Data driven biases - class imbalance24:02 - Bias within the features27:09 - Mitigate biases in the model/dataset33:20 - Automated debiasing from learned latent structure37:11 - Adaptive latent space debiasing39:39 - Evaluation towards decreased racial and gender bias41:00 - Summary and future considerations for AI fairnessSubscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
Co-Learning of Task and Sensor Placement for Soft Robotics (Teaser) Alexander Amini 2021-03-22 | Unlike rigid robots which operate with compact degrees of freedom, soft robots must reason about an infinite dimensional state space. Mapping this continuum state space presents significant challenges, especially when working with a finite set of discrete sensors. Reconstructing the robot’s state from these sparse inputs is challenging, especially since sensor location has a profound downstream impact on the richness of learned models for robotic tasks. In this work, we present a novel representation for co-learning sensor placement and complex tasks. Specifically, we present a neural architecture which processes on-board sensor information to learn a salient and sparse selection of placements for optimal task performance. We evaluate our model and learningalgorithm on six soft robot morphologies for various supervised learning tasks, including tactile sensing and proprioception. We also highlight applications to soft robot motion subspace visualization and control. Our method demonstrates superior performance in task learning to algorithmic and human baselines while also learning sensor placements and latent spaces that are semantically meaningful.Authors: Andrew Spielberg*, Alexander Amini*, Lillian Chin, Wojciech Matusik, and Daniela RusPublished in: IEEE Robotics and Automation Letters (RA-L), with presentation in RoboSoft 2021.OpenAccess (free) paper link: ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=9345345Full paper video: youtube.com/watch?v=cffniAE1RRk
MIT 6.S191: Evidential Deep Learning and Uncertainty Alexander Amini 2021-03-19 | MIT Introduction to Deep Learning 6.S191: Lecture 7Evidential Deep Learning and Uncertainty EstimationLecturer: Alexander AminiJanuary 2021For all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline0:00 - Introduction and motivation5:00 - Outline for lecture5:50 - Probabilistic learning8:33 - Discrete vs continuous target learning14:12 - Likelihood vs confidence17:40 - Types of uncertainty21:15 - Aleatoric vs epistemic uncertainty22:35 - Bayesian neural networks28:55 - Beyond sampling for uncertainty31:40 - Evidential deep learning33:29 - Evidential learning for regression and classification42:05 - Evidential model and training45:06 - Applications of evidential learning46:25 - Comparison of uncertainty estimation approaches47:47 - ConclusionSubscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191 (2021): Deep Learning New Frontiers Alexander Amini 2021-03-12 | MIT Introduction to Deep Learning 6.S191: Lecture 6Deep Learning Limitations and New FrontiersLecturer: Ava SoleimanyJanuary 2021For all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline0:00 - Introduction1:11 - Course logistics3:48 - Upcoming hot topics and guest lectures6:56 - Deep learning and expressivity of NNs10:02 - Generalization of deep models14:03 - Neural network failure modes 18:43 - Uncertainty in deep learning22:41 - Adversarial attacks26:35 - Algorithmic bias27:27 - Limitations summary28:29 - Structure in DL 29:46 - Learning on graphs37:50 - Learning on 3D pointclouds39:58 - Automated Machine Learning (AutoML)48:03 - ConclusionSubscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191 (2021): Reinforcement Learning Alexander Amini 2021-03-05 | MIT Introduction to Deep Learning 6.S191: Lecture 5Deep Reinforcement LearningLecturer: Alexander AminiJanuary 2021For all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline0:00 - Introduction3:17 - Classes of learning problems6:19 - Definitions12:33 - The Q function16:14 - Deeper into the Q function20:49 - Deep Q Networks26:28 - Atari results and limitations29:53 - Policy learning algorithms33:11 - Discrete vs continuous actions37:22 - Training policy gradients44:50 - RL in real life46:02 - VISTA simulator47:44 - AlphaGo and AlphaZero and MuZero55:22 - SummarySubscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191 (2021): Deep Generative Modeling Alexander Amini 2021-02-26 | MIT 6.S191 (2021): Introduction to Deep LearningDeep Generative ModelingLecturer: Ava SoleimanyJanuary 2021For all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline0:00 - Introduction6:03 - Why care about generative models? 8:56 - Latent variable models11:31 - Autoencoders17:00 - Variational autoencoders24:30 - Priors on the latent distribution34:38 - Reparameterization trick38:14 - Latent perturbation and disentanglement 41:25 - Debiasing with VAEs43:42 - Generative adversarial networks46:14 - Intuitions behind GANs48:27 - Training GANs52:57 - GANs: Recent advances57:15 - CycleGAN of unpaired translation1:01:01 - SummarySubscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191 (2021): Convolutional Neural Networks Alexander Amini 2021-02-19 | MIT Introduction to Deep Learning 6.S191: Lecture 3Convolutional Neural Networks for Computer VisionLecturer: Alexander AminiJanuary 2021For all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline0:00 - Introduction2:47 - Amazing applications of vision7:56 - What computers "see"14:02 - Learning visual features18:50 - Feature extraction and convolution22:20 - The convolution operation27:27 - Convolution neural networks34:05 - Non-linearity and pooling38:59 - End-to-end code example40:25 - Applications42:02 - Object detection50:52 - End-to-end self driving cars54:00 - SummarySubscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191 (2021): Recurrent Neural Networks Alexander Amini 2021-02-12 | MIT Introduction to Deep Learning 6.S191: Lecture 2Recurrent Neural NetworksLecturer: Ava SoleimanyJanuary 2021For all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline0:00 - Introduction2:37 - Sequence modeling4:54 - Neurons with recurrence12:07 - Recurrent neural networks14:13 - RNN intuition17:01 - Unfolding RNNs18:39 - RNNs from scratch22:12 - Design criteria for sequential modelling23:37 - Word prediction example31:31 - Backpropagation through time33:40 - Gradient issues38:46 - Long short term memory (LSTM)47:47 - RNN applications52:15 - Attention59:24 - SummarySubscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191 (2021): Introduction to Deep Learning Alexander Amini 2021-02-05 | MIT Introduction to Deep Learning 6.S191: Lecture 1Foundations of Deep LearningLecturer: Alexander AminiFor all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline0:00 - Introduction4:48 - Course information10:18 - Why deep learning?12:28 - The perceptron14:42 - Activation functions17:48 - Perceptron example21:43 - From perceptrons to neural networks27:42 - Applying neural networks 30:21 - Loss functions33:23 - Training and gradient descent38:05 - Backpropagation43:06 - Setting the learning rate47:17 - Batched gradient descent49:49 - Regularization: dropout and early stopping55:55 - SummarySubscribe to stay up to date with new deep learning lectures at MIT, or follow us on @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT Deep Learning 6.S191 Teaser Alexander Amini 2021-01-04 | MIT Introduction to Deep Learning: 6.S191.Massachusetts Institute of Technology (MIT)*New 2021 Edition Coming Soon!*For more details: http://introtodeeplearning.comMIT'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!Subscribe here on Youtube stay up to date with new deep learning lectures at MIT, or follow us on @MITDeepLearning on Twitter and Instagram to stay *fully-connected*!!Credits: - Alexander Amini and Ava Soleimany (course instructors)- Max Kessler (course footage)- Larry Zhang (campus drone footage)
Learning Autonomous Driving in Simulation Alexander Amini 2020-09-02 | A TechCrunch spotlight feature covering a brand new data-driven simulation engine, developed at MIT, capable of training full-scale autonomous vehicles in simulation for deployment in the real-world. This project represented the first time a virtual vehicle, trained using reinforcement learning, was able to learn entirely in simulation and be successfully deployed onto real roads. To learn more about our simulator (VISTA) and this project:Project website: http://bit.ly/VISTA-simTechnical Paper: ieeexplore.ieee.org/document/8957584Code: coming soon!! https://forms.gle/4CzEou4Ffs9Lg9dr7
Learning Robust Control Policies for End-to-End Driving in Simulation | RA-L/ICRA 2020 Alexander Amini 2020-06-09 | This talk is streamed as part of a presentation for the IEEE International Conference on Robotics and Automation (ICRA) on the paper entitled: Learning Robust Control Policies for End-to-End Autonomous Driving From Data-Driven Simulation.Amini, A., Gilitschenski, I., Phillips, J., Moseyko, J., Banerjee, R., Karaman, S., & Rus, D. (2020). IEEE Robotics and Automation Letters, 5(2), 1143-1150.To get more details on this work, read the paper, or access the code please visit: http://www.mit.edu/~amini/vista/Abstract:In this work, we present a data-driven simulation and training engine capable of learning end-to-end autonomous vehicle control policies using only sparse rewards. By leveraging real, human-collected trajectories through an environment, we render novel training data that allows virtual agents to drive along a continuum of new local trajectories consistent with the road appearance and semantics, each with a different view of the scene. We demonstrate the ability of policies learned within our simulator to generalize to and navigate in previously unseen real-world roads, without access to any human control labels during training. Our results validate the learned policy onboard a full-scale autonomous vehicle, including in previously un-encountered scenarios, such as new roads and novel, complex, near-crash situations. Our methods are scalable, leverage reinforcement learning, and apply broadly to situations requiring effective perception and robust operation in the physical world.
MIT 6.S191 (2020): Machine Learning for Scent Alexander Amini 2020-04-10 | MIT Introduction to Deep Learning 6.S191: Lecture 10Machine Learning for ScentLecturer: Alex Wiltschko (Google Brain)January 2020For all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline0:00 - Introduction2:55 - Digitizing smell4:28 - The sense of smell10:11 - Problem setup12:58 - Molecule fragrance dataset16:00 - Baseline algorithms18:14 - Graph neural networks21:25 - Molecules to graphs23:03 - Predicting odor descriptors25:19 - The odor embedding space27:58 - Molecular neighbors30:04 - Generalization32:34 - Explaining/interpreting predictions36:49 - Summary and future workSubscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191 (2020): Neural Rendering Alexander Amini 2020-04-03 | MIT Introduction to Deep Learning 6.S191: Lecture 9Neural RenderingLecturer: Chuan Li (Lambda Labs)January 2020For all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline0:00 - Introduction5:40 - Forward rendering12:18 - End-to-end rendering14:20 - 3D data representations16:12 - RenderNet (Voxels)21:00 - Neural point based graphics (Pointclouds)24:06 - Mesh model rendering25:00 - Inverse rendering28:33 - HoloGAN34:40 - SummarySubscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191 (2020): Generalizable Autonomy for Robot Manipulation Alexander Amini 2020-03-27 | MIT Introduction to Deep Learning 6.S191: Lecture 8Generalizable Autonomy for Robot ManipulationLecturer: Animesh Garg (NVIDIA & University of Toronto)January 2020For all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline0:00 - Introduction3:45 - Achieving generalizable autonomy4:19 - Leveraging imitation learning6:08 - Learning visuo-motor policies13:09 - Learning skills16:38 - Off-policy RL + AC-Teach22:02 - Compositional planning27:20 - Model-based RL 34:37 - Leveraging task structure36:35 - Neural task programming (NTP)43:04 - Data for robotics44:24 - RoboTurk45:54 - SummarySubscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191 (2020): Neurosymbolic AI Alexander Amini 2020-03-20 | MIT Introduction to Deep Learning 6.S191: Lecture 7Neurosymbolic Hybrid Artificial IntelligenceLecturer: David CoxJanuary 2020For all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline0:00 - Introduction1:25 - Evolution of AI7:32 - MIT-IBM Watson AI Lab10:10 - Why is AI today "narrow"?19:17 - Out-of-distribution performance21:07 - ObjectNet23:04 - Adversarial examples25:24 - When does deep learning struggle?27:00 - Neural networks vs symbolic AI28:38 - Neurosymbolic AI34:20 - Advantages of combining symbolic AI39:01 - CLEVERER and more40:40 - SummarySubscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191 (2020): Deep Learning New Frontiers Alexander Amini 2020-03-13 | MIT Introduction to Deep Learning 6.S191: Lecture 6Deep Learning Limitations and New FrontiersLecturer: Ava SoleimanyJanuary 2020For all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline0:00 - Introduction0:58 - Course logistics3:59 - Upcoming guest lectures5:35 - Deep learning and expressivity of NNs10:02 - Generalization of deep models14:14 - Adversarial attacks17:00 - Limitations summary18:18 - Structure in deep learning22:53 - Uncertainty & bayesian deep learning28:09 - Deep evidential regression33:08 - AutoML36:43 - ConclusionSubscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT Autonomous Vehicle: Learning Robust Sim-to-Real Control Policies Alexander Amini 2020-03-10 | Learning Robust Control Policies for End-to-End Autonomous Driving From Data-Driven Simulation(The VISTA Simulation Engine)In this work, we present a data-driven simulation and training engine capable of learning end-to-end autonomous vehicle control policies using only sparse rewards. By leveraging real, human-collected trajectories through an environment, we render novel training data that allows virtual agents to drive along a continuum of new local trajectories consistent with the road appearance and semantics, each with a different view of the scene. We demonstrate the ability of policies learned within our simulator to generalize to and navigate in previously unseen real-world roads, without access to any human control labels during training. Our results validate the learned policy onboard a full-scale autonomous vehicle, including in previously un-encountered scenarios, such as new roads and novel, complex, near-crash situations. Our methods are scalable, leverage reinforcement learning, and apply broadly to situations requiring effective perception and robust operation in the physical world.This paper was published in IEEE Robotics and Automation Letters and will be presented at ICRA 2020.Technical Paper: ieeexplore.ieee.org/document/8957584Code: coming soon!! https://forms.gle/4CzEou4Ffs9Lg9dr7Project website: http://bit.ly/VISTA-simAuthors: Alexander Amini, Igor Gilitschenski, Jacob Phillips, Julia Moseyko, Rohan Banerjee, Sertac Karaman, Daniela RusAcknowledgments: Support for this work was given by the National Science Foundation (NSF) and Toyota Research Institute (TRI). However, note that this article solely reflects the opinions and conclusions of its authors and not TRI or any other Toyota entity. We gratefully acknowledge the support of NVIDIA Corporation with the donation of the V100 GPU and Drive PX2 used for this research.
MIT 6.S191 (2020): Reinforcement Learning Alexander Amini 2020-03-06 | MIT Introduction to Deep Learning 6.S191: Lecture 5Deep Reinforcement LearningLecturer: Alexander AminiJanuary 2020For all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline0:00 - Introduction2:47 - Classes of learning problems4:59 - Definitions9:23 - The Q function13:18 - Deeper into the Q function17:17 - Deep Q Networks21:44 - Atari results and limitations24:13 - Policy learning algorithms27:36 - Discrete vs continuous actions30:11 - Training policy gradients36:04 - RL in real life37:40 - VISTA simulator38:55 - AlphaGo and AlphaZero42:51 - SummarySubscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191 (2020): Deep Generative Modeling Alexander Amini 2020-02-28 | MIT 6.S191 (2020): Introduction to Deep LearningDeep Generative ModelingLecturer: Ava SoleimanyJanuary 2020For all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline0:00 - Introduction4:37 - Why do we care? 6:36 - Latent variable models8:12 - Autoencoders13:30 - Variational autoencoders20:18 - Reparameterization trick23:55 - Latent pertubation26:12 - Debiasing with VAEs30:40 - Generative adversarial networks32:40 - Intuitions behind GANs35:12 - GANs: Recent advances39:38 - SummarySubscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191 (2020): Convolutional Neural Networks Alexander Amini 2020-02-21 | MIT Introduction to Deep Learning 6.S191: Lecture 3Convolutional Neural Networks for Computer VisionLecturer: Alexander AminiJanuary 2020For all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline0:00 - Introduction3:04 - What computers "see"8:06 - Learning visual features12:36 - Feature extraction and convolution19:12 - Convolution neural networks24:03 - Non-linearity and pooling28:30 - Code example29:32 - Applications32:53 - End-to-end self driving cars35:55 - SummarySubscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191 (2020): Recurrent Neural Networks Alexander Amini 2020-02-14 | MIT Introduction to Deep Learning 6.S191: Lecture 2Recurrent Neural NetworksLecturer: Ava SoleimanyJanuary 2020For all lectures, slides, and lab materials: http://introtodeeplearning.comLecture Outline0:00 - Introduction2:39 - Sequence modeling9:57 - Recurrent neural networks14:04 - RNN intuition16:48 - Unfolding RNNs20:31 - Backpropagation through time24:32 - Gradient issues28:57 - Long short term memory (LSTM)37:36 - RNN applications41:30 - Attention44:05 - SummarySubscribe to stay up to date with new deep learning lectures at MIT, or follow us @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
Barack Obama: Intro to Deep Learning | MIT 6.S191 Alexander Amini 2020-02-10 | MIT Introduction to Deep Learning 6.S191 (2020)DISCLAIMER: The following video is synthetic and was created using deep learning with simultaneous speech-to-speech translation as well as video dialogue replacement (CannyAI). ** NOTE**: The audio quality demonstrated here was additionally degraded since we want to avoid improper use of this technology. The purpose of this video is to excite the class about the potential of deep learning, not to deceive anyone. Thus, we purposely lowered the audio quality before publishing to make the synthetic aspect of this video clearer. All course lectures are posted online: youtube.com/watch?v=njKP3FqW3Sk&list=PLtBw6njQRU-rwp5__7C0oIVt26ZgjG9NI&index=1For all lectures, slides, and lab materials vist http://introtodeeplearning.comSubscribe to stay up to date with new deep learning lectures at MIT, or follow us on @MITDeepLearning on Twitter and Instagram to stay fully-connected!!
MIT 6.S191 (2019): Visualization for Machine Learning (Google Brain) Alexander Amini 2019-03-20 | MIT Introduction to Deep Learning 6.S191: Lecture 7Data Visualization for Machine LearningLecturer: Fernanda ViegasGoogle Brain Guest LectureJanuary 2019For all lectures, slides and lab materials: http://introtodeeplearning.com
