Deep learning 1.0 and Beyond, Part 1
2 likes552 views
The document presents a comprehensive tutorial on deep learning, addressing its evolution into deep learning 2.0, focusing on various models including transformers and graph neural networks. It explores key concepts such as attention mechanisms, neural architecture search, and unsupervised learning methods, detailing their benefits and applications. Additionally, the tutorial emphasizes the scalability and adaptability of deep learning models across various domains and tasks.
Deep learning 1.0 and Beyond, Part 1
- 1. 16/11/2020 1 A/Prof Truyen Tran With contribution from Vuong Le, Hung Le, Thao Le, Tin Pham & Dung Nguyen Deakin University December 2020 Deep learning 1.0 and Beyond A tutorial Part I @truyenoz truyentran.github.io truyen.tran@deakin.edu.au letdataspeak.blogspot.com goo.gl/3jJ1O0 linkedin.com/in/truyen-tran
- 2. 2012 2016 AusDM 2016 Turing Awards 2018 GPT-3 2020 8 years snapshot
- 3. Why (still) DL? Practical Generality: Applicable to many domains. Competitive: DL is hard to beat as long as there are data to train. Scalability: DL is better with more data, and it is very scalable. Theoretical Expressiveness: Neural nets can approximate any function. Learnability: Neural nets are trained easily. Generalisability: Neural nets generalize surprisingly well to unseen data.
- 4. It is easy to get lost in current DL zoo 16/11/2020 4 Vietnam News AAAI’20
- 6. Model design goals Resource adaptive, compressible Easy to train Use (almost) no labels Ability to extrapolate Support both fast and slow learning Support both fast and slow inference 16/11/2020 6 Uniformity Universality Scalability Reusability Capture long-term dependencies in time and space Capture invariances natively
- 7. Neural memories Theory of mind Neural reasoning A system view Deep learning 2.0 16/11/2020 7 Agenda Classic models Transformers Graph neural networks Unsupervised learning Deep learning 1.0
- 8. Deep models via layer stacking Theoretically powerful, but limited in practice Integrate-and-fire neuron andreykurenkov.com Feature detector Block representation16/11/2020 8
- 9. http://torch.ch/blog/2016/02/04/resnets.html Practice Shorten path length with skip-connections Easier information and gradient flows 16/11/2020 9 http://qiita.com/supersaiakujin/items/935bbc9610d0f87607e8 Theory
- 10. Sequence model with recurrence Assume the stationary world Classification Image captioning Sentence classification Neural machine translation Sequence labelling Source: http://karpathy.github.io/assets/rnn/diags.jpeg 16/11/2020 10
- 11. Spatial model with convolutions Assume filters/motifs are translation invariant http://colah.github.io/posts/2015-09-NN-Types-FP/ Learnable kernels andreykurenkov.com Feature detector, often many
- 12. Convolutional networks Summarizing filter responses, destroying locations adeshpande3.github.io 16/11/2020 12
- 13. Operator on sets/bags: Attentions Not everything is created equal for a goal Need attention model to select or ignore certain computations or inputs Can be “soft” (differentiable) or “hard” (requires RL) Attention provides a short-cut long- term dependencies Also encourages sparsity if done right! http://distill.pub/2016/augmented-rnns/
- 14. Fast weights | HyperNet The world is recursive Early ideas in early 1990s by Juergen Schmidhuber and collaborators. Data-dependent weights | Using a controller to generate weights of the main net. 16/11/2020 14 Ha, David, Andrew Dai, and Quoc V. Le. "Hypernetworks." arXiv preprint arXiv:1609.09106 (2016).
- 15. Neural architecture search When design is cheap and non-creative The space is huge and discrete Can be done through meta-heuristics (e.g., genetic algorithms) or Reinforcement learning (e.g., one discrete change in model structure is an action). 16/11/2020 15 Bello, Irwan, et al. "Neural optimizer search with reinforcement learning." arXiv preprint arXiv:1709.07417 (2017).
- 16. Neural memories Theory of mind Neural reasoning A system view Deep learning 2.0 16/11/2020 16 Agenda Classic models Transformers Graph neural networks Unsupervised learning Deep learning 1.0
- 17. Motivations RNN is theoretically powerful, but purely sequential, hence slow and has limited effective memory for finite size. Augmenting with external memories solve some problem, but still slow CNN is a feed-forward net, can be parallelized, but theoretically not too strong – random long-term dependencies are hard to encode Prior to 2017, most architectures are mixture of FNN, RNN and CNN Non- uniformity, hard to scale to a large number of tasks. We need supports for Parallel computation Long-rang dependency encoding (constant path length) Uniform construction (e.g., like columnar structure of neocortex) 16/11/2020 17
- 18. Prelim: Memory networks Input is a set Load into memory, which is NOT updated. State is a RNN with attention reading from inputs Concepts: Query, key and content + Content addressing. Deep models, but constant path length from input to output. Equivalent to a RNN with shared input set. 16/11/2020 18 Sukhbaatar, Sainbayar, Jason Weston, and Rob Fergus. "End-to-end memory networks." Advances in neural information processing systems. 2015.
- 19. Transformers: The triumph of self-attention 16/11/2020 19 Tay, Yi, et al. "Efficient transformers: A survey." arXiv preprint arXiv:2009.06732 (2020). State KeyQuery Memory
- 20. Transformers are (new) Hopfield net 16/11/2020 20 Ramsauer, Hubert, et al. "Hopfield networks is all you need." arXiv preprint arXiv:2008.02217 (2020).
- 21. Transformer v.s. memory networks Memory network: Attention to input set One hidden state update at a time. Final state integrate information of the set, conditioned on the query. Transformer: Loading all inputs into working memory Assigns one hidden state per input element. All hidden states (including those from the query) to compute the answer. 16/11/2020 21
- 22. Universal transformers 16/11/2020 22 https://ai.googleblog.com/2018/08/moving-beyond-translation-with.html Dehghani, Mostafa, et al. "Universal Transformers." International Conference on Learning Representations. 2018.
- 23. Efficient Transformers Transformer is quadratic in time Cannot deal with large sets (or sequence) 16/11/2020 23 Tay, Yi, et al. "Efficient transformers: A survey." arXiv preprint arXiv:2009.06732 (2020).
- 24. Neural memories Theory of mind Neural reasoning A system view Deep learning 2.0 16/11/2020 24 Agenda Classic models Transformers Graph neural networks Unsupervised learning Deep learning 1.0
- 25. Why graphs? Graphs are pervasive in many scientific disciplines. Deep learning needs to move beyond vector, fixed-size data. The sub-area of graph representation has reached a certain maturity, with multiple reviews, workshops and papers at top AI/ML venues. 16/11/2020 25 NeurIPS 2020
- 27. Biology, pharmacy & chemistry Molecule as graph: atoms as nodes, chemical bonds as edges Computing molecular properties Chemical-chemical interaction Chemical reaction 16/11/2020 27 #REF: Penmatsa, Aravind, Kevin H. Wang, and Eric Gouaux. "X- ray structure of dopamine transporter elucidates antidepressant mechanism." Nature 503.7474 (2013): 85-90. Gilmer, Justin, et al. "Neural message passing for quantum chemistry." arXiv preprint arXiv:1704.01212 (2017).
- 28. Materials science 16/11/2020 28 Xie, Tian, and Jeffrey C. Grossman. "Crystal Graph Convolutional Neural Networks for an Accurate and Interpretable Prediction of Material Properties." Physical review letters 120.14 (2018): 145301. • Crystal properties • Exploring/generating solid structures • Inverse design
- 29. Videos as space-time region graphs (Abhinav Gupta et al, ECCV’18)
- 30. Basic neural graph mechanism: Message passing 16/11/2020 30 #REF: Pham, Trang, et al. "Column Networks for Collective Classification." AAAI. 2017. Relation graph GCN update rule, vector form GCN update rule, matrix form Generalized message passing
- 31. Attention: Not all messages are created equal (Do et al arXiv’s17, Veličković et al ICLR’ 18) 16/11/2020 31 Learning deep matrix representations, K Do, T Tran, S Venkatesh, arXiv preprint arXiv:1703.01454
- 32. Neural graph morphism Input: Graph Output: A new graph. Same nodes, different edges. Model: Graph morphism Method: Graph transformation policy network (GTPN) 16/11/2020 32 Kien Do, Truyen Tran, and Svetha Venkatesh. "Graph Transformation Policy Network for Chemical Reaction Prediction." KDD’19.
- 33. Neural graph recurrence Graphs that represent interaction between entities through time Spatial edges are node interaction at a time step Temporal edges are consistency relationship through time
- 34. ASSIGN: Asynchronous, Sparse Interaction Graph Network (Morais et al, 2021 @ A2I2, Deakin – Work in Progress) 16/11/2020 35
- 35. Graph generation No regular structures (e.g. grid, sequence,…) Graphs are permutation invariant: #permutations are exponential function of #nodes The probability of a generated graph G need to be marginalized over all possible permutations Generating graphs with variable size Aim for diversity of generated graphs
- 36. Generation methods Classical random graph models, e.g., An exponential family of probability distributions for directed graphs (Holland and Leinhardt, 1981) Deep generative models: GraphVAE, Graphphite, Junction Tree VAE, GAN variants etc. Sequence-based & RL methods 16/11/2020 37
- 37. GraphRNN A case of graph dynamics: nodes and edges are added sequentially. Solve tractability using BFS 16/11/2020 38 You, Jiaxuan, et al. "GraphRNN: Generating realistic graphs with deep auto-regressive models." ICML (2018).
- 38. Graphs step-wise construction using reinforcement learning 16/11/2020 39 You, Jiaxuan, et al. "Graph Convolutional Policy Network for Goal-Directed Molecular Graph Generation." NeurIPS (2018). Graph rep (message passing) | graph validation (RL) | graph faithfulness (GAN)
- 39. Neural memories Theory of mind Neural reasoning A system view Deep learning 2.0 16/11/2020 41 Agenda Classic models Transformers Graph neural networks Unsupervised learning Deep learning 1.0
- 40. Unsupervised learning 16/11/2020 42 Photo credit: Brandon/Flickr Humans mainly learn by exploring without clear instructions and labelling
- 41. Representation learning, a bit of history “Representation is the use of signs that stand in for and take the place of something else” It has been a goal of neural networks since the 1980s and the current wave of deep learning (2005-present) Replacing feature engineering Between 2006-2012, many unsupervised learning models with varying degree of success: RBM, DBN, DBM, DAE, DDAE, PSD Between 2013-2018, most models were supervised, following AlexNet Since 2018, unsupervised learning has become competitive (with contrastive learning, self-supervised learning, BERT)! 16/11/2020 43
- 43. Criteria for a good representation Separates factors of variation (aka disentanglement), which are linearly correlated with desired outputs of downstream tasks. Provides abstraction that is invariant against deformations and small variations. Is distributed (one concept is represented by multiple units), which is compact and good for interpolation. Optionally, offers dimensionality reduction. Optionally, is sparse, giving room for emerging symbols. 16/11/2020 45 Bengio, Yoshua, Aaron Courville, and Pascal Vincent. "Representation learning: A review and new perspectives." IEEE transactions on pattern analysis and machine intelligence 35.8 (2013): 1798-1828.
- 44. Why neural unsupervised learning? Neural nets have representational richness: FFN are functional approximator RNN are program approximator, can estimate a program behaviour and generate a string CNN are for translation invariance Transformers are powerful contextual encoder Compactness: Representations are (sparse and) distributed. Essential to perception, compact storage and reasoning Accounting for uncertainty: Neural nets can be stochastic to model distributions Symbolic representation: realisation through sparse activations and gating mechanisms 16/11/2020 46
- 45. Neural autoregressive models: Predict the next step given the history The keys: (a) long-term dependencies, (b) ordering, & (c) parameter sharing. Can be realized using: RNN CNN: One-sided CNN, dilated CNN (e.g., WaveNet), PixelCNN Transformers GPT-X family Masked autoencoder MADE Pros: General, good quality thus far Cons: Slow – needs better inductive biases for scalability16/11/2020 47 lyusungwon.github.io/studies/2018/07/25/nade/
- 46. Generative models: Discover the underlying process that generates data 16/11/2020 48 Many applications: • Text to speech • Simulate data that are hard to obtain/share in real life (e.g., healthcare) • Generate meaningful sentences conditioned on some input (foreign language, image, video) • Semi-supervised learning • Planning
- 47. Deep (Denoising) AutoEncoder: Self-reconstruction of data 16/11/2020 49 Auto-encoderFeature detector Representation Raw data (optionally with added noise) Reconstruction Deep Auto-encoder Encoder Decoder
- 48. Credit: kvfrans.com Gaussian hidden variables Data Generative net Recognising net Variational Autoencoder Approximating the posterior by a neural net Two separate processes: generative (hidden visible) versus recognition (visible hidden)
- 49. GAN: Generative Adversarial nets Matching data statistics Yann LeCun: GAN is one of best idea in past 10 years! Instead of modeling the entire distribution of data, learns to map ANY random distribution into the region of data, so that there is no discriminator that can distinguish sampled data from real data. Any random distribution in any space Binary discriminator, usually a neural classifier Neural net that maps z x
- 50. Generative adversarial networks (Adapted from Goodfellow’s, NIPS 2014) 16/11/2020 52
- 51. Progressive GAN: Generated images 16/11/2020 53 Karras, T., Aila, T., Laine, S., & Lehtinen, J. (2017). Progressive growing of gans for improved quality, stability, and variation. arXiv preprint arXiv:1710.10196.
- 52. BERT Transformer that predicts its own masked parts BERT is like parallel approximate pseudo- likelihood ~ Maximizing the conditional likelihood of some variables given the rest. When the number of variables is large, this converses to MLE (maximum likelihood estimate). 16/11/2020 54 https://towardsdatascience.com/bert-explained-state-of-the-art-language-model-for-nlp-f8b21a9b6270
- 53. Contrastive learning: Comparing samples 16/11/2020 55 Le-Khac, Phuc H., Graham Healy, and Alan F. Smeaton. "Contrastive Representation Learning: A Framework and Review." arXiv preprint arXiv:2010.05113 (2020).
- 54. Unsupervised learning: A few more points No external labels, but rich training signals (thousand bits per sample, as opposed to a few bits in supervised learning). A few techniques: Compressing data as much as possible with little loss Energy-based, i.e., pull down energy of observed data, pull up every else Filling the missing slots (aka predictive learning, self-supervised learning) We have not covered unsupervised learning on graphs (e.g., DeepWalk, GPT-GNN), but the general principles should hold. Question: Multiple objectives, or no objective at all? Question: Emergence from many simple interacting elements? 16/11/2020 56 Liu, Xiao, et al. "Self-supervised learning: Generative or contrastive." arXiv preprint arXiv:2006.08218 (2020).
- 55. End of part I 16/11/2020 57