deep learning for dialog
play

Deep Learning for Dialog Nate Kushman Researcher Microsoft - PowerPoint PPT Presentation

Sep 15, 2023 •314 likes •915 views

Deep Learning for Dialog Nate Kushman Researcher Microsoft Research Labs Microsoft Research Labs Cambridge, UK Basic Computer Science Research Long-term goal: real-world impact

  1. Deep Learning for Dialog Nate Kushman Researcher Microsoft Research Labs

  2. Microsoft Research Labs Cambridge, UK  Basic Computer Science Research  Long-term goal: real-world impact

  3. https://www.microsoft.com/en-us/research/blog/microsoft-unveils-project-brainwave/

  4. Example problem from Gaunt et al. 2017 STRUCTURE PROGRAM LEARNING MACHINE synthesising source code to guide a traditional program descent to synthesize human- Can we use neural networks Machine learning for How can we use gradient interpretable programs? synthesizer?

  5. Representation Learning: Generative Models 𝑑 𝐻 𝑡 𝑗 𝛸 𝑑 𝛸 𝑡 𝜄 𝑦 𝑗 𝑗 ∈ 𝐻 𝐻 ∈ G Multi-Level VAE

  6. observation action Project Malmo:

  7. Agent Applications Services Infrastructure Optimal medication outcomes require a concert of patient, practitioner and health-system insights and actions, in more timely and targeted ways: higher definition healthcare

  8. ech Services?  Seamless integration between human and AI agents Example: Customer Care Intelligence  Frictionless human-like conversations Why is Dialog Relevant for Fin/T

  15. 2016 2015 2014 2013 Speech 2012 2011 2010 2009

  16. 2016 2015 2014 2013 Speech 2012 2011 2010 2009

  17. 2016 2015 2014 Vision 2013 Speech 2012 2011 2010 2009

  18. 2016 2015 Natural Language 2014 2013 2012 Vision 2011 Speech 2010 2009

  19. • Manual engineering for Large amounts of data Many Possible Combinations each new domain Need Either: • “The button is above the menu” “The menu is below the button” Followed by: “Now Click it” Subtlety Matters Context is challenging for three reasons: Three main Challenges: Solution: “Upgrade VS240 driver” Long Distance “I’m printing in powerpoint ” Relationships ” “I run Windows 10” …. “

  20. Struggle with long distance relationships ” Neural Context Representations 9.1 In practice used mostly for chit-chat Requires large amounts of data 8.9 9.0 8.3 2.7 6.2 Excel at Subtlety Symbolic Context Representations Great for long distance relationships Dominant approach in Real-World Systems Requires Engineering per domain Struggle with subtly of meaning Intent: Projector Setup Domain: Technical Support Device: Epson VS240 “

  21. 0 0 0 1 T 1 0 0 0 A 0 1 0 0 G 0 0 1 0 G C 0 1 0 0 1 0 0 0 A task: Classify junk or gene? 0 0 0 1 data: DNA sequences T 1 0 0 0 A 0 1 0 0 G Example 0 0 1 0 G C 0 1 0 0

  22. 0 0 0 1 T 1 0 0 0 A H t r t H t+1 0 1 0 0 G 0 0 1 0 H t – “an h - dimensional ‘compressed summary’ of t tokens” G C 0 1 0 0 t + 2 NN 1 0 0 0 H t+1 A t + 1 0 0 0 1 T t 1 0 0 0 A t - 1 0 1 0 0 G H t h 0 0 1 0 G C 0 1 0 0

  23. 0 0 0 1 T 1 0 0 0 A 0 1 0 0 G 0 0 1 0 G C NN 0 1 0 0 t + 2 H t+2 1 0 0 0 A t + 1 0 0 0 1 T t 1 0 0 0 A t - 1 0 1 0 0 G 0 0 1 0 G C 0 1 0 0

  24. 0 0 0 1 T 1 0 0 0 A 0 1 0 0 G 0 0 1 0 G C NN 0 1 0 0 t + 2 H t+2 NN 1 0 0 0 A t + 1 0 0 0 1 T t 1 0 0 0 A t - 1 0 1 0 0 G H t 0 0 1 0 G C 0 1 0 0

  25. 0 0 0 1 T 1 0 0 0 A 0 1 0 0 G 0 0 1 0 G C NN 0 1 0 0 t + 2 H t+2 NN 1 0 0 0 A t + 1 NN 0 0 0 1 T t NN 1 0 0 0 A t - 1 NN 0 1 0 0 G NN 0 0 1 0 G C NN 0 1 0 0

  26. 0 0 0 0 0 1 … … … [10 2 – 10 3 ] [10 4 – 10 5 ] 1 0 0 0 Aardwolf 0 1 0 0 0 0 0 0 … Aardvark Zymurgy 0 0 ... ... x t 0 1 0 0 embedding 1 0 0 0 0 1 0 0 0 0 1 0 0 0 0 1 G A C T

  27. great great <end> That’s am happy <start> That’s 3. Sequence-to-sequence I 2. Next-token prediction <end> am happy am happy I 1. Sequence classification am happy  I

  28. <end> Output: Next Utterance Decoder That’s great <start> Context Neural Input: Preceding dialog context am happy Encoder I

  29. Supervised Learning Faster learning but requires a human to perform the task to provide the “correct” response <end> Output: Next Utterance Decoder That’s great <start> Context Neural Input: Preceding dialog context am happy Encoder I

  30. Reinforcement Learning +10 10 -10 Hard to learn from, but only requires users to provide a rating at the end of a dialog <end> Output: Next Utterance Decoder You’re mean <start> Context Neural Input: Preceding dialog context am happy Encoder I

  31. Datatbase Domain: Movie Are there any action movies Language Intent: Find Slot - Genre: “Action” Understanding to see this weekend? Slot - Date: “this weekend” Dialog Managment Natural Where would Language Query(LOCATION) you like to go? Generation Domain: Movie Intent: Find Slot - Genre: “Action” Slot - Date: “this weekend” Language Dialog How about the Capital Theater? Slot - Location: “Capital Theater” Understanding Management Domain: Movie Intent: Find Slot - Genre: “Action” Slot - Date: “this weekend” Slot – Location: “Capital”

  32. Slot - Date: “this weekend” Slot - Type: “Taiwanese” Slot - Genre: “Action” Slot – Loc : “Oakland” Slot - Price: “cheap” Domain: Restaurant Intent: Find Domain: Movie Intent: Find Understanding Understanding Language Language Are there any action movies Find me a cheap Taiwanese Resturant in Oakland? to see this weekend? • Pipeline Decision • Ontology Based aurant nt in in Domain Domain Movie Restaura

  33. 2,5% RNN Error Rate Domain: Movie Traditional (ngram) 9,5% Domain: Movie weekend? Are there any action movies to see this weekend? Are there

  34. Slot - Date: “this weekend” Slot - Genre: “Action” 3,4% RNN Error Rate Domain: Movie Intent: Find Traditional 4,3% (SVM) this weekend? B-Date I-Date see <> <> to B-Genre I-Genre action movies Are there any action movies Domain: Movie to see this weekend? any <> there <> Are <>

  35. Joint RNN Slot – Date: “this weekend” 13,4% Slot – Genre: “Action” Error Rate Separate RNN Intent: Find Domain: Movie 13,7% Intent: Find Domain: Movie <EOS> Understanding this weekend? B-Date I-Date Language see <> <> to B-Genre I-Genre action movies Are there any action movies to see this weekend? any <> there <> Are <>

  36. Domain: Movie Are there any action movies Language Intent: Find Slot - Genre: “Action” Understanding to see this weekend? Slot - Date: “this weekend” Dialog Managment Natural Where would Language Query(LOCATION) you like to go? Generation Domain: Movie Intent: Find Slot - Genre: “Action” Slot - Date: “this weekend” Language Dialog How about the Capital Theater? Slot - Location: “Capital Theater” Understanding Management Domain: Movie Intent: Find Slot - Genre: “Action” Slot - Date: “this weekend” Slot – Location: “Capital”

  37. Slot - Date: “this weekend” Slot - Date: “this weekend” Slot – Location: “Capital” Slot - Genre: “Action” Slot - Genre: “Action” Management Dialog Domain: Movie Intent: Find Domain: Movie Intent: Find Slot - Location: “Capital Theater” Query(LOCATION)

  38. Domain: Movie Are there any action movies Language Intent: Find Slot - Genre: “Action” Understanding to see this weekend? Slot - Date: “this weekend” Dialog Managment Natural Where would Language Query(LOCATION) you like to go? Generation Domain: Movie Intent: Find Slot - Genre: “Action” Slot - Date: “this weekend” Language Dialog How about the Capital Theater? Slot - Location: “Capital Theater” Understanding Management Domain: Movie Intent: Find Slot - Genre: “Action” Slot - Date: “this weekend” Slot – Location: “Capital”

  39. Query(LOCATION) go? to like Generation Language Natural would you Where <S> Where would you like to go?

  40. Management Domain: Movie Managment Domain: Movie Dialog Dialog … … Domain … Query… Slot… Understanding Understanding Generation Language Language Language Natural How about the Capital Theater? you like to go? Are there any action movies Where would to see this weekend?

  41. Management Domain: Movie Managment Domain: Movie Dialog Dialog … … Domain … Query… Slot… Utterance Encoder Utterance Encoder Generation Language Natural How about the Capital Theater? you like to go? Are there any action movies Where would to see this weekend?

  42. Domain: Movie Domain: Movie … … Domain … Query… Slot… Utterance Encoder Utterance Encoder Generation Language Natural How about the Capital Theater? you like to go? Are there any action movies Where would to see this weekend?

  43. Domain: Movie Domain: Movie … … Domain … Query… Slot… Utterance Decoder Utterance Encoder Utterance Encoder How about the Capital Theater? you like to go? Are there any action movies Where would to see this weekend?

  44. Dialog Encoder Context Context Neural Neural Context Context Neural Neural Utterance Decoder Utterance Encoder Utterance Encoder How about the Capital Theater? you like to go? Are there any action movies Where would to see this weekend?

Recommend

Replika Building an Emotional conversation with Deep Learning Replika: History Luka Luka

Replika Building an Emotional conversation with Deep Learning Replika: History Luka Luka

Replika Building an Emotional conversation with Deep Learning Replika: History Luka Luka Replika Restaurant Personality bots: Your AI friend recommendations Prince, Roman Dialog Architecture Typical scenario: Small talk Dialog

748 views • 26 slides
Advanced NLU &amp; Dialog Models Ling575 Spoken Dialog Systems April 21, 2016 Roadmap

Advanced NLU &amp; Dialog Models Ling575 Spoken Dialog Systems April 21, 2016 Roadmap

Advanced NLU &amp; Dialog Models Ling575 Spoken Dialog Systems April 21, 2016 Roadmap Advanced NLU Advanced Dialog Models Information State Models Statistical Dialog Models Learning Probabilistic Slot Filling

627 views • 60 slides
Hao Su July 6, 2017 Outline Overview of 3D deep learning 3D deep learning algorithms

Hao Su July 6, 2017 Outline Overview of 3D deep learning 3D deep learning algorithms

Deep 3D Representation Learning for Visual Computing Hao Su July 6, 2017 Outline Overview of 3D deep learning 3D deep learning algorithms Conclusion 2 Outline Overview of 3D deep learning Background 3D deep learning tasks 3D deep

1.66k views • 122 slides
ACCELERATE DEEP LEARNING WITH NVIDIA'S DEEP LEARNING PLATFORM | STEPHEN JONES | GTC16 DEEP

ACCELERATE DEEP LEARNING WITH NVIDIA'S DEEP LEARNING PLATFORM | STEPHEN JONES | GTC16 DEEP

ACCELERATE DEEP LEARNING WITH NVIDIA'S DEEP LEARNING PLATFORM | STEPHEN JONES | GTC16 DEEP LEARNING EVERYWHERE INTERNET &amp; CLOUD MEDICINE &amp; BIOLOGY MEDIA &amp; ENTERTAINMENT SECURITY &amp; DEFENSE AUTONOMOUS MACHINES Image

308 views • 26 slides
Deep Learning on GPUs March 2016 What is Deep Learning? GPUs and DL AGENDA DL in practice

Deep Learning on GPUs March 2016 What is Deep Learning? GPUs and DL AGENDA DL in practice

Deep Learning on GPUs March 2016 What is Deep Learning? GPUs and DL AGENDA DL in practice Scaling up DL 2 What is Deep Learning? 3 DEEP LEARNING EVERYWHERE INTERNET &amp; CLOUD MEDICINE &amp; BIOLOGY SECURITY &amp; DEFENSE MEDIA &amp;

1.45k views • 39 slides
Reproducibility and Replicability in Deep Reinforcement Learning (and Other Deep Learning

Reproducibility and Replicability in Deep Reinforcement Learning (and Other Deep Learning

Reproducibility and Replicability in Deep Reinforcement Learning (and Other Deep Learning Methods) Peter Henderson Statistical Society of Canada Annual Meeting 2018 Contributors Riashat Islam Phil Bachman Doina Precup David Meger Joelle

850 views • 37 slides
SDS: ASR, NLU, &amp; VXML Ling575 Spoken Dialog April 14, 2016 Roadmap Dialog System

SDS: ASR, NLU, &amp; VXML Ling575 Spoken Dialog April 14, 2016 Roadmap Dialog System

SDS: ASR, NLU, &amp; VXML Ling575 Spoken Dialog April 14, 2016 Roadmap Dialog System components: ASR: Noisy channel model Representation Decoding NLU: Call routing Grammars for dialog systems Basic

925 views • 76 slides
Image Classification with DIGITS NVIDIA Deep Learning Institute 1 DEEP LEARNING INSTITUTE DLI

Image Classification with DIGITS NVIDIA Deep Learning Institute 1 DEEP LEARNING INSTITUTE DLI

Image Classification with DIGITS NVIDIA Deep Learning Institute 1 DEEP LEARNING INSTITUTE DLI Mission Helping people solve challenging problems using AI and deep learning. Developers, data scientists and engineers Self-driving cars,

1.44k views • 72 slides
Deep Neural Networks and Deep Reinforcement Learning Deep Learning, Goodfellow, Bengio and

Deep Neural Networks and Deep Reinforcement Learning Deep Learning, Goodfellow, Bengio and

Deep Neural Networks and Deep Reinforcement Learning Deep Neural Networks and Deep Reinforcement Learning Deep Learning, Goodfellow, Bengio and Courville [chapt. 6,7,8]; AIMA [sect. 21.1-21.3]; Sutton and Barto, Reinforcement Learning: an

527 views • 35 slides
Natural Language Processing with Deep Learning CS224N The Future of Deep Learning + NLP Kevin

Natural Language Processing with Deep Learning CS224N The Future of Deep Learning + NLP Kevin

Natural Language Processing with Deep Learning CS224N The Future of Deep Learning + NLP Kevin Clark Deep Learning for NLP 5 years ago No Seq2Seq No Attention No large-scale QA/reading comprehension datasets No TensorFlow or

1.41k views • 92 slides
An Overview of Deep Residual Learning Semih Yagcioglu 01.03.2016 Deep Residual Learning

An Overview of Deep Residual Learning Semih Yagcioglu 01.03.2016 Deep Residual Learning

An Overview of Deep Residual Learning Semih Yagcioglu 01.03.2016 Deep Residual Learning Microsoft Research Asia (MSRA) Kaiming He, Xiangyu Zhang, Shaoqing Ren, &amp; Jian Sun. Deep Residual Learning for Image Recognition. arXiv

1.32k views • 54 slides
Deep Reinforcement Learning [Mastering the Game of Go with Deep Reinforcement Learning and Tree

Deep Reinforcement Learning [Mastering the Game of Go with Deep Reinforcement Learning and Tree

Deep Reinforcement Learning [Mastering the Game of Go with Deep Reinforcement Learning and Tree Search, Nature 2016] CS 486/686 University of Waterloo Lecture 21: July 12, 2017 Outline AlphaGo Supervised Learning of Policy Networks

541 views • 15 slides
Mocha.jl Deep Learning in Julia Chiyuan Zhang (@pluskid) CSAIL, MIT Deep Learning Learning

Mocha.jl Deep Learning in Julia Chiyuan Zhang (@pluskid) CSAIL, MIT Deep Learning Learning

Mocha.jl Deep Learning in Julia Chiyuan Zhang (@pluskid) CSAIL, MIT Deep Learning Learning with multi-layer (3~30) neural networks, on a huge training set. State-of-the-art on many AI tasks Computer Vision : Image Classification,

1.53k views • 23 slides
Minjie Wang Deep Learning Deep Learning trend in the past 10 years Caffe State-of-art DL

Minjie Wang Deep Learning Deep Learning trend in the past 10 years Caffe State-of-art DL

Tofu: Parallelizing Deep Learning Systems with Automatic Tiling Minjie Wang Deep Learning Deep Learning trend in the past 10 years Caffe State-of-art DL system is based on dataflow GPU#0 w1 w2 data g1 g2 Forward propagation

405 views • 38 slides
Deep Learning: Theory and Practice Deep Learning - Practical 02-04-2020 Considerations

Deep Learning: Theory and Practice Deep Learning - Practical 02-04-2020 Considerations

Deep Learning: Theory and Practice Deep Learning - Practical 02-04-2020 Considerations deeplearning.cce2020@gmail.com Deep Networks Intuition Neural networks with multiple hidden layers - Deep networks [Hinton, 2006] Deep Networks Intuition

1.33k views • 28 slides
Deep Learning: State of the Art (2020) Deep Learning Lecture Series https://deeplearning.mit.edu

Deep Learning: State of the Art (2020) Deep Learning Lecture Series https://deeplearning.mit.edu

Deep Learning: State of the Art (2020) Deep Learning Lecture Series https://deeplearning.mit.edu 2020 For the full list of references visit: http://bit.ly/deeplearn-sota-2020 Outline Deep Learning Growth, Celebrations, and Limitations

1.1k views • 87 slides
12. Unsupervised Deep Learning CS 535 Deep Learning, Winter 2018 Fuxin Li With materials from

12. Unsupervised Deep Learning CS 535 Deep Learning, Winter 2018 Fuxin Li With materials from

12. Unsupervised Deep Learning CS 535 Deep Learning, Winter 2018 Fuxin Li With materials from Wanli Ouyang, Zsolt Kira, Lawrence Neal, Raymond Yeh, Junting Lou and Teck-Yian Lim Unsupervised Learning in General Unsupervised learning is

1.48k views • 115 slides
Regularization for Deep Learning Lecture slides for Chapter 7 of Deep Learning

Regularization for Deep Learning Lecture slides for Chapter 7 of Deep Learning

Regularization for Deep Learning Lecture slides for Chapter 7 of Deep Learning www.deeplearningbook.org Ian Goodfellow 2016-09-27 Definition Regularization is any modification we make to a learning algorithm that is intended to reduce

179 views • 13 slides
Introduction to Deep Learning Outline Deep Learning RNN CNN Attention

Introduction to Deep Learning Outline Deep Learning RNN CNN Attention

Introduction to Deep Learning Outline Deep Learning RNN CNN Attention Transformer Pytorch Introduction Basics Examples RNNs Some slides borrowed from Fei-Fei Li &amp; Justin Johnson &amp;

1.1k views • 55 slides
Dialog as a Vehicle for Lifelong Learning Aishwarya Padmakumar, Raymond J. Mooney Department of

Dialog as a Vehicle for Lifelong Learning Aishwarya Padmakumar, Raymond J. Mooney Department of

Dialog as a Vehicle for Lifelong Learning Aishwarya Padmakumar, Raymond J. Mooney Department of Computer Science The University of Texas at Austin Standard Supervised Learning Pipeline Collect Train Test Labelled Model Model Data

745 views • 17 slides
Deep learning for NLP: Introduction CS 6956: Deep Learning for NLP Words are a very fantastical

Deep learning for NLP: Introduction CS 6956: Deep Learning for NLP Words are a very fantastical

Deep learning for NLP: Introduction CS 6956: Deep Learning for NLP Words are a very fantastical banquet, just so many strange dishes And yet, we seem to do fine We can understand and generate language effortlessly. Almost 1 Wouldnt it be

1.62k views • 52 slides
Deep Reinforcement Learning and Complex Environments Raia Hadsell End-to-end Deep Learning

Deep Reinforcement Learning and Complex Environments Raia Hadsell End-to-end Deep Learning

Deep Reinforcement Learning and Complex Environments Raia Hadsell End-to-end Deep Learning for robots? slide from V. Vanhoucke End-to-end Deep Learning for robots? 2010 : Speech Recognition Audio Acoustic Model Phonetic Model

1.05k views • 65 slides
End-to-end LSTM-based dialog control optimized with supervised and reinforcement learning

End-to-end LSTM-based dialog control optimized with supervised and reinforcement learning

End-to-end LSTM-based dialog control optimized with supervised and reinforcement learning Authors: Jason D. Williams and Geoffrey Zweig Speaker: Hamidreza Shahidi Outline Introduction Model description Optimizing with

388 views • 28 slides
Bayesian Deep Learning Mohd Adnan Problems With Deep Learning What does a model not know?

Bayesian Deep Learning Mohd Adnan Problems With Deep Learning What does a model not know?

Bayesian Deep Learning Mohd Adnan Problems With Deep Learning What does a model not know? Uninterpretable black-boxes Easily fooled (AI safety) Lacks solid mathematical foundation Crucially relies on big dat Why

297 views • 15 slides

More recommend