All Segments
Deep Learning State of the Art (2020)
51 segments available
Lecture on most recent research and developments in deep learning, and hopes for 2020. This is not intended to be a list of SOTA benchmark results, but rather a set of highlights of machine learning and AI innovations and progress in academia, industry, and society in general. This lecture is part of the MIT Deep Learning Lecture Series. Website: https://deeplearning.mit.edu Slides: http://bit.ly/2QEfbAm References: http://bit.ly/deeplearn-sota-2020 Playlist: http://bit.ly/deep-learning-playlist OUTLINE: 0:00 - Introduction 0:33 - AI in the context of human history 5:47 - Deep learning celebrations, growth, and limitations 6:35 - Deep learning early key figures 9:29 - Limitations of deep learning 11:01 - Hopes for 2020: deep learning community and research 12:50 - Deep learning frameworks: TensorFlow and PyTorch 15:11 - Deep RL frameworks 16:13 - Hopes for 2020: deep learning and deep RL frameworks 17:53 - Natural language processing 19:42 - Megatron, XLNet, ALBERT 21:21 - Write with transformer examples 24:28 - GPT-2 release strategies report 26:25 - Multi-domain dialogue 27:13 - Commonsense reasoning 28:26 - Alexa prize and open-domain conversation 33:44 - Hopes for 2020: natural language processing 35:11 - Deep RL and self-play 35:30 - OpenAI Five and Dota 2 37:04 - DeepMind Quake III Arena 39:07 - DeepMind AlphaStar 41:09 - Pluribus: six-player no-limit Texas hold'em poker 43:13 - OpenAI Rubik's Cube 44:49 - Hopes for 2020: Deep RL and self-play 45:52 - Science of deep learning 46:01 - Lottery ticket hypothesis 47:29 - Disentangled representations 48:34 - Deep double descent 49:30 - Hopes for 2020: science of deep learning 50:56 - Autonomous vehicles and AI-assisted driving 51:50 - Waymo 52:42 - Tesla Autopilot 57:03 - Open question for Level 2 and Level 4 approaches 59:55 - Hopes for 2020: autonomous vehicles and AI-assisted driving 1:01:43 - Government, politics, policy 1:03:03 - Recommendation systems and policy 1:05:36 - Hopes for 2020: Politics, policy and recommendation systems 1:06:50 - Courses, Tutorials, Books 1:10:05 - General hopes for 2020 1:11:19 - Recipe for progress in AI 1:14:15 - Q&A: what made you interested in AI 1:15:21 - Q&A: Will machines ever be able to think and feel? 1:18:20 - Q&A: Is RL a good candidate for achieving AGI? 1:21:31 - Q&A: Are autonomous vehicles responsive to sound? 1:22:43 - Q&A: What does the future with AGI look like? 1:25:50 - Q&A: Will AGI systems become our masters? CONNECT: - If you enjoyed this video, please subscribe to this channel. - Twitter: https://twitter.com/lexfridman - LinkedIn: https://www.linkedin.com/in/lexfridman - Facebook: https://www.facebook.com/lexfridman - Instagram: https://www.instagram.com/lexfridman
Segments Timeline
The Dream of AI: A Historical Perspective
This segment introduces the lecture by reflecting on the historical context of artificial intelligence, tracing its roots back to humanity's ancient desire to create intelligent machines. The speaker emphasizes the significance of understanding the human brain and the journey of AI from its inception to the present day, highlighting key milestones and figures in the field.
"welcome to 2020 and welcome to the deep learning lecture series let's start it off today to take a quick whirlwind tour of all the exciting things that happened in seventeen eighteen and nineteen espe..."
Celebrating Deep Learning: Turing Award and Key Figures
In this segment, the speaker celebrates the achievements in deep learning, including the Turing Award awarded to pioneers like Geoffrey Hinton, Yann LeCun, and Yoshua Bengio. The discussion covers the evolution of neural networks, the contributions of early figures like Frank Rosenblatt, and the importance of recognizing the broader community that has shaped deep learning.
"development since the early modern human being is when we've seen a lot of the machinery Machin was born not in stories but in actuality is the machine was engineered since the Industrial Revolution a..."
The Limitations of Deep Learning
This segment addresses the limitations of deep learning, discussing the skepticism surrounding its capabilities and the challenges it faces in achieving true artificial intelligence. The speaker highlights the need for critical evaluation and the importance of recognizing the boundaries of current technologies while still pushing for progress.
"folks have received the award but of course the community that contributed to deep learning is bigger much bigger than those three many of whom might be here today at MIT broadly in academia in indust..."
Hopes for 2020: Growth and Collaboration
The speaker shares hopes for the future of deep learning in 2020, emphasizing the need for interdisciplinary collaboration and open-mindedness in research. This segment outlines the importance of addressing common sense reasoning, active learning, and ethical considerations in AI development.
"community but not too much like a little spice in the soup of progress aside from that kind of skepticism the growth of cvpr iclear europe's all these conference submission papers has grown year over ..."
Deep Learning Frameworks: TensorFlow and PyTorch
This segment discusses the maturation of deep learning frameworks, focusing on TensorFlow and PyTorch. The speaker highlights the convergence of features between these frameworks and their impact on the community, making deep learning more accessible to beginners and researchers alike.
"flaws in our data and the flaws and our human ethics and then robotics in terms of deep learning application robotics I'd love to see a lot of development continued development deep reinforcement lear..."
Reinforcement Learning Frameworks: Current Landscape
In this segment, the speaker reviews the current landscape of reinforcement learning frameworks, recommending stable baselines for beginners. The discussion includes the evolution of various libraries and the importance of clear documentation and ease of use in advancing the field.
"goodbye cruel world okay on the reinforcement learning front we're kind of in the same space as JavaScript libraries are in there's no clear winners coming out if if you're a beginner in the space the..."
Future Aspirations: Framework Agnosticism
The speaker expresses hopes for framework agnostic research in 2020, advocating for easier model transfer between TensorFlow and PyTorch. This segment emphasizes the need for collaboration and standardization in deep learning frameworks to enhance usability across disciplines.
"that we can all agree on much like opening idea for the environment world has done and continued work that Kerris has started and many other rappers around tensorflow started of greater and greater ab..."
Innovations in Transformer Models
The speaker introduces several notable transformer models, including Megatron, XLNet, and ALBERT. He explains how these models build upon BERT's architecture, achieving state-of-the-art results in NLP tasks while also discussing their unique features and contributions to the field.
"results on a lot of language benchmarks from synthesis classification to tagging question answering and so on there's hundreds of data sets and benchmarks that emerged most of which Burt has dominated..."
The Power of Pre-trained Models
In this segment, the speaker highlights the significance of pre-trained transformer models and the tools available for utilizing them. He mentions Hugging Face as a key player in making these models accessible, and discusses the implications of their widespread use in NLP applications.
"Nvidia huge transformer a few tools have emerged so one on hugging face is a company and also a repository that has implemented in both pi torch intensive flow or a lot of these transformer based nati..."
XLNet: The New King of Transformers
The speaker discusses XLNet, a model that combines the strengths of BERT and recurrent neural networks to achieve superior performance on various NLP tasks. He explains its innovative approach to embeddings and how it has set new benchmarks in the field.
"in a parallel way model and data parallelism in the training the first breakthrough results in terms of performance the model that replaced Bert as king of transformers is XL net from CMU of Google re..."
Understanding Limitations of Language Models
This segment addresses the limitations of current language models, particularly in terms of understanding and reasoning. The speaker shares insights on how models like GPT-2 generate text and the philosophical implications of their capabilities, emphasizing the distinction between memorization and true understanding.
"state-of-the-art results on 12 an LP tasks including the the difficult Stanford question answering benchmark of squad 2 and they provide that provide open source tensorflow implementation including a ..."
Common Sense Reasoning Challenges
The speaker explores the challenges of common sense reasoning in AI, discussing the need for hybrid systems that integrate symbolic reasoning with deep learning. He highlights recent research efforts aimed at improving AI's ability to understand and reason about everyday concepts.
"learn before this yesterday actually just came up with a bunch of prompts so on the left is a prompt you give it the meaning of life here for example is not what I think it is it's what I do to make i..."
The Importance of Dialogue Systems
In this segment, the speaker emphasizes the significance of dialogue systems in AI, particularly in multi-domain conversations. He discusses the challenges faced in developing effective dialogue systems and the need for advancements in state tracking and context management.
"is probably Andrew and I have to agree so this model knows what it's doing and I tried to get it to say something nice about me and that's a lot of attempts so this is kind of funny is finally did it ..."
Ethical Considerations in AI Release
The speaker reflects on the ethical implications of releasing powerful AI models, using GPT-2 as a case study. He discusses the potential risks associated with misinformation and the importance of responsible communication between AI developers and the public.
"with blocks you can ask it about gravity all those kinds of things it shows that it doesn't understand the fundamentals of the concepts that are being reasoned about and I'll mention of work that take..."
The Role of Machine Learning in Conversations
This segment examines the role of machine learning in enhancing conversational AI. The speaker discusses the limitations of current systems in achieving meaningful dialogue and the need for more sophisticated approaches to understanding context and intent.
"think while it turned out that the GPG to model is not quite so dangerous that humans are in fact more dangerous than AI currently the that thought experiment is very interesting they released a repor..."
Advancements in Multi-Domain Dialogue
The speaker highlights recent advancements in multi-domain dialogue systems, discussing the challenges of maintaining context and coherence across different topics. He shares insights from recent research and the potential for improving conversational AI.
"difficult because we as the public seem to penalize anybody trying to have that conversation and the model of sharing privately confidentially between ml machine learning organizations and experts is ..."
Common Sense in AI Conversations
In this segment, the speaker discusses the integration of common sense reasoning into AI conversations. He presents examples of how AI can struggle with basic reasoning tasks and the importance of developing systems that can understand and respond to everyday scenarios.
"five domain challenging very difficult fide domain human to human dialogue dataset there's a few ideas there I should probably hurry up and start skipping stuff the common sense reasoning which is rea..."
Lessons from Open-Domain Conversations
The speaker shares lessons learned from open-domain conversation systems, emphasizing the need for engaging and entertaining interactions. He discusses the importance of maintaining user interest and the challenges faced in creating compelling dialogue.
"language model would come in is that usually a hamburger with friends indicates a good time so you basically take the question generate the common sense concept and from that be able to determine the ..."
The Future of Conversational AI
In this concluding segment, the speaker reflects on the future of conversational AI, discussing the challenges and opportunities that lie ahead. He emphasizes the need for continued innovation and the importance of creating systems that can engage users meaningfully.
"few lessons from Alcoa that I particularly like and this is kind of echoes the work in the IBM Watson who the Jeopardy challenge is that one of the big ones is that machine learning is not an essentia..."
The Art of Conversation in AI
This segment explores the nuances of natural language systems and their limitations in expressing opinions. The speaker emphasizes that conveying intelligence often requires being opinionated and confident, highlighting the importance of entertainment in conversations. The discussion touches on the Turing Test and the Lobner Prize, setting benchmarks for conversational AI.
"beauty the humour the wit the fun of conversations you jump jump around from topic to topic and opinions one of the things that natural language systems don't seem to have much is opinions if I learne..."
Context Switching in Conversations
The speaker discusses the challenges of maintaining context in natural language conversations, illustrating how successful interactions often involve jumping between topics. They provide examples of conversational AI systems and their limitations in context retention, emphasizing the need for systems that can navigate multi-domain dialogues effectively.
"okay lots of lessons to learn there this is really the lobner prize the Turing test of our generation that's I'm excited to see if there's anybody able to solve the lexer prize again a lexer prize is ..."
Advancements in Natural Language Processing
This segment highlights the advancements in natural language processing (NLP) and the hopes for 2020. The speaker discusses the importance of common-sense reasoning and the ability of language models to maintain context over longer texts. They reference transformer models like XLNet and the potential for self-supervised learning to enhance NLP capabilities.
"how conversation goes change you to change in and back quickly there's been a lot of sequins to sequins kind of work using natural language to summarize a lot of applications one of the for me I clear..."
Transformers and Visual Information
The discussion shifts to the application of transformer models beyond text, exploring the hope for 2020 to transfer their success to visual information. The speaker emphasizes the potential for deep reinforcement learning (DRL) and self-play in enhancing AI's understanding of the world through video and visual data.
"context which transformers again with excel net transformer Excel is starting to be able to do but we're still far away from that long-term lifelong maintenance of context dialogue open domain dialogu..."
Reinforcement Learning in Gaming
This segment delves into the exciting developments in reinforcement learning (RL) within gaming, particularly focusing on OpenAI's Dota 2 project. The speaker discusses the significant computational resources required for training and the concept of self-play, where AI systems improve by competing against themselves.
"information the world of video for example DRL and self play this has been an exciting year continues to be an exciting time for reinforcement learning in games and robotics so first dota2 an open AI ..."
Multi-Agent Learning and Collaboration
The speaker discusses the complexities of multi-agent learning, where individual agents must learn to cooperate and compete. They highlight the philosophical implications of collective intelligence and the potential for RL to explore social behaviors that emerge in both AI and human interactions.
"because the because of the natural process of self play that's a fascinating process the 2019 version the last version of open AI 5 well has a 99.9 win rate versus the 2018 version ok then deep mind a..."
AlphaStar and StarCraft II
This segment focuses on DeepMind's AlphaStar and its achievements in StarCraft II, illustrating how AI can learn and adapt in complex environments. The speaker emphasizes the significance of using real-world observations to train AI, showcasing the potential for AI to reach grandmaster levels in competitive gaming.
"human to humans social systems okay here's some visualizations the agents automatically figure out as you see in other games they figure out the concepts so knowing very little knowing nothing about t..."
Reinforcement Learning in Poker
The discussion shifts to the application of reinforcement learning in poker, specifically the Pluribus AI developed by CMU. The speaker explains how self-play and Monte Carlo methods were used to create a competitive poker player capable of defeating professionals, highlighting the strategic complexities involved.
"encourage you to observe a lot of the interesting on their blog posts and videos of the different strategies that the there are our Allegiance are able to figure out here's a quote from the one of the..."
Robotic Manipulation and the Rubik's Cube
This segment explores the advancements in robotic manipulation through reinforcement learning, particularly in solving the Rubik's Cube. The speaker discusses the use of automatic domain randomization to create challenging environments for training, emphasizing the potential for AI to learn complex tasks through iterative processes.
"on the imperfect information game side poker in 2018 CMU no Brown I was able to beat had two head-to-head No Limit Texas Hold'em and now team six player No Limit Texas Hold'em against professional pla..."
Emergent Learning in Robotics
The speaker discusses the concept of emergent meta-learning in robotics, where neural networks adapt to increasingly difficult tasks. They express hopes for 2020 regarding the exploration of social behaviors in multi-agent systems and the potential for reinforcement learning to provide insights into human behavior.
"thin value bets on the river he's very good at extracting value out of his good hands sort of making bets without scaring off the opponent Darren Elias said it's major strength is its ability to use m..."
The Lottery Ticket Hypothesis
This segment introduces the lottery ticket hypothesis, which suggests that smaller sub-networks within larger neural networks can achieve similar performance. The speaker discusses the implications of this hypothesis for network efficiency and the potential for discovering more effective architectures in deep learning.
"psychology department one day like where you use reinforcement learning to study to reverse-engineer human behavior and study it through that way and again in games I'm not sure with the big challenge..."
Disentangled Representations in Learning
This segment delves into the concept of disentangled representations, where each part of a vector learns distinct concepts about a dataset. The speaker discusses the challenges of achieving this without inductive biases and highlights the importance of explicit biases in learning effective representations.
"is disentangle representations which again to serve its own lecture but here showing a 10 vector representation and the goal is where each part of the vector can learn one particular concept about a d..."
Understanding Deep Double Descent
The speaker explains the phenomenon of deep double descent, where increasing the number of parameters in a neural network initially increases test error before decreasing again. This segment discusses the implications of this behavior for model training dynamics and the ongoing research questions surrounding it.
"of the exciting is the double dissent idea that's been extended and to the deep you know network context by open AI to explore the the phenomena that as we increase the number of parameters in a neura..."
Advancements in Autonomous Vehicles
This segment highlights the exciting developments in autonomous vehicles, focusing on the two approaches: Level 2, where human supervision is required, and Level 4, where AI takes full responsibility. The speaker discusses the progress made by companies like Waymo and Tesla in real-world applications of AI in driving.
"deep learning in 2020 is to continue exploring the fundamentals of model selection training dynamics the folks focused on the performance of the training in terms of memory and speed has worked on and..."
Waymo vs. Tesla: Approaches to Autonomy
The speaker contrasts the approaches of Waymo and Tesla in developing autonomous vehicles. Waymo's extensive testing and simulation strategies are discussed alongside Tesla's reliance on deep learning for perception and action, emphasizing the different methodologies in achieving autonomy.
"vehicles oh boy let me try to use this few sentences as possible to describe this section of a few slides it is one of the most exciting areas of applications of AI and learning in the real world toda..."
The Role of Active Learning in AI
This segment explores the concept of active learning in AI, where neural networks continuously learn from edge cases and improve over time. The speaker emphasizes the importance of iterative learning processes and the need for companies to adopt these methods for effective machine learning applications.
"the exciting thing is that there is seven hundred thousand eight hundred thousand Tesla autopilot systems that means there's these systems that are human supervised they're using fun a multi-headed ne..."
Challenges in Autonomous Driving
The speaker discusses the challenges faced in autonomous driving, including the need for human vigilance and the complexities of perception and action. This segment highlights the ongoing debates about the effectiveness of different approaches and the critical nature of safety in autonomous systems.
"to improve until it's brilliant and that process is specially interesting when you take it outside of single task learning so most papers are written on single task learning you take whatever benchmar..."
Comparing Sensor Technologies for Autonomy
This segment compares the sensor technologies used in autonomous vehicles, such as cameras, lidar, and radar. The speaker discusses the pros and cons of each approach, emphasizing the importance of data quality and the challenges of integrating multiple sensor types for reliable performance.
"camera based systems have the highest resolution so that it's very amenable to learning but the con is that it requires a lot of data a huge amount of data and when nobody knows how much data yet the ..."
Future Directions in Autonomous Vehicle Learning
The speaker expresses hopes for future advancements in autonomous vehicle learning, particularly in active and multitask learning. This segment emphasizes the need for continuous improvement and the importance of deploying updates to enhance the capabilities of autonomous systems.
"watching movies so on and so on the things that people naturally do the open question is how good can auto pilot get before that becomes a serious problem and if that decrement nullifies the safety be..."
The Importance of Public Datasets
This segment discusses the necessity of public datasets for training autonomous vehicles, particularly for edge cases. The speaker calls for more transparency and collaboration among automotive companies to share data and improve the development of AI in driving.
"applied deep learning innovation like I mentioned these are really exciting areas at least to me of active learning multitask learning and lifelong learning online learning iterative learning there's ..."
AI in Politics: A New Frontier
The speaker briefly touches on the intersection of AI and politics, highlighting the growing discussions around artificial intelligence in political contexts. This segment reflects on the implications of AI in governance and the need for informed dialogue on its impact.
"companies and simulators Carla and video draft constellation voyage deep drive there's a bunch of simulators coming out that are allowing people to experiment with perception with planning with reinfo..."
The Power of Recommendation Systems
The speaker delves into the significance of recommendation systems powered by deep learning algorithms. They discuss how these systems shape our information consumption and communication, urging for transparency and ethical considerations in their development. The segment emphasizes the societal impact of these algorithms and the need for engineers to be aware of their implications.
"these early developments early ideas from the from the federal government about what what are the dangers and what are the hopes the the funding and the education required to build a successful infras..."
Learning from AI: Courses and Resources
In this segment, the speaker recommends various courses and resources for learning about deep learning and reinforcement learning. They highlight the importance of hands-on experience and suggest specific courses from Fast.ai, Coursera, and Stanford. The speaker encourages aspiring learners to engage with practical tutorials and foundational texts to deepen their understanding of AI.
"ideas and also I believe it's the the the role of companies to publish more has been very little published on the details of recommendation systems behind Twitter Facebook YouTube Google so all those ..."
The Role of Active Learning in AI
The speaker emphasizes the importance of active learning in the real-world application of deep learning. They express a desire to see more research in this area, highlighting its potential to improve AI systems over time. The segment discusses the need for AI to continually learn from experiences, paralleling human learning processes.
"love to see more of that so my hope in this in the politic space in the public discourse space for 2020 is less fear of AI and more discourse between government and experts on topics of privacy cybers..."
Challenges and Critiques of Deep Learning
This segment addresses the skepticism surrounding deep learning and its limitations. The speaker acknowledges the importance of criticism in moderation and emphasizes perseverance in the field. They reflect on the historical context of AI development and the need for innovative thinking to overcome current challenges.
"is controlling how we think how we see the world the moral system under which we operate quickly to mention and wrapping up with a few minutes of questions if there are any is the deep learning course..."
Building the Future of AI
The speaker shares their personal journey into AI, discussing their early inspirations and the evolution of their interests. They reflect on the potential for machines to think and feel, suggesting that while AI may simulate emotions, the ethical implications of such developments must be carefully considered.
"the mathematics non linear algebra statistics and I have a few lectures up online that you should never watch then on the reinforcement learning side David silver is one of the greatest people in unde..."
The Spark of AI Inspiration
Lex Fridman shares his initial inspiration for pursuing AI, reflecting on his early aspirations to become a psychiatrist. He discusses how he transitioned from wanting to engineer the human mind to building artificial intelligence, emphasizing the importance of programming in understanding and creating intelligent systems.
"Lex I'm wondering if you recall what was the initial spark or inspiration that drove you towards work in AI was it when you were pretty young or was it in more recent years so I want to become a psych..."
Will Machines Think and Feel?
In this segment, Lex Fridman confidently asserts that machines will eventually be able to think and feel emotions. He explores the concept of machines faking emotions and thoughts, using humorous examples of Roomba vacuums to illustrate how emotional displays can be perceived by humans.
"understand the - to build it speaking of building mind do you personally think that machines will ever be able to think and the second question will they ever be able to feel emotions a hundred percen..."
Ethics of AI Emotions
Fridman delves into the ethical implications of AI potentially experiencing emotions and suffering. He discusses the observer's role in defining suffering and the significance of AI systems expressing emotions, suggesting that the first instance of an AI claiming to suffer will raise serious ethical questions.
"that's the sort of everything else is everything else is impossible to pin down I'm asking so what about the ethical aspects of it I'm asking because I was born in the Soviet Union as well and one of ..."
Reinforcement Learning and AGI
In this thought-provoking segment, Lex Fridman discusses whether reinforcement learning is a viable path to achieving artificial general intelligence (AGI). He emphasizes the importance of real-world interaction and common sense reasoning, suggesting that traditional reinforcement learning may not be sufficient on its own.
"and I can do we can do that today like I already built the Roombas I they won't sell currently but I think the first time a Roomba says please don't hurt me that's when we start to have serious conver..."
The Future of AI and Human Interaction
Fridman speculates on the future of AI and its role in human society. He discusses the potential for AI to serve as companions and the ethical considerations surrounding rights for AI entities. He draws parallels between societal views on animals and future perspectives on intelligent machines.
"gaps are I think the ability to but I'm so human centric but I think the approach of being able to take knowledge and put it together sort of building into more and more complicated pieces of informat..."
AI as Tools of Power
In this concluding segment, Lex Fridman warns about the potential dangers of AI being controlled by a few powerful individuals or corporations. He emphasizes the need for democratizing AI technology to prevent misuse and ensure that AI serves humanity rather than becoming a tool for oppression.
"about the autonomous vehicles whether they are responsive to environmental sounds I mean like notice in her cart autonomous vehicle driving erratically won't respond to my beep that's a really interes..."