Memory advances in Neural Turing Machines
3 likes479 views
This document discusses research on improving neural Turing machines through the use of program memory. It introduces the neural universal Turing machine (NUTM), which augments neural Turing machines with a neural stored-program memory (NSM) to store programs. The NSM allows NUTMs to sequence tasks, do continual learning, and answer questions - addressing limitations of current neural Turing machines. Further research opportunities are outlined, including using memory for graphs and relational structures, memory-supported reasoning, and developing full cognitive architectures.
![Problems of current NTMs
Lack of theoretical analysis on optimal memory operations.
Previous works are based on intuitions:
Location-based reading/writing; temporal linkage reading; least-used
writing [Santoro et.al, Graves et.al]
Sparse access over big memory [Rae et.al]
Very slow due to heavy memory read/write computations
12](https://image.slidesharecdn.com/vinai-ntm-june-2019-191027004426/85/Memory-advances-in-Neural-Turing-Machines-12-320.jpg)
Memory advances in Neural Turing Machines
- 1. 8/06/2019 1 Hanoi, June 2019 Truyen Tran Deakin University @truyenoz truyentran.github.io truyen.tran@deakin.edu.au letdataspeak.blogspot.com goo.gl/3jJ1O0 Memory Advances in Neural Turing Machines
- 2. 8/06/2019 2 Deep Learning Domain expert Knowledge-based
- 3. 8/06/2019 3 Can we learn from data a model that is as powerful as a Turing machine? In other words, can we learn a (neural) program that learns to program from data?
- 4. 8/06/2019 4 Program memory Outlook Sparse read/write Variational memory Neural Turing Machine Agenda
- 5. Modelling Three interwoven processes: • Disease progression • Interventions & care processes • Recording rules Example: Electronic medical records 8/06/2019 5 Source: medicalbillingcodings.org visits/admissions time gap ? prediction point Abstraction Need memory to handle thousands of events, compute complex healthcare “grammars”, support chain of reasoning, rapid switching of tasks.
- 6. Neural Turing machine (NTM) A controller that takes input/output and talks to an external memory module. Memory has read/write operations. The main issue is where to write, and how to update the memory state. All operations are differentiable. https://rylanschaeffer.github.io/content/research/neural_turing_machine/main.html
- 7. 8/06/2019 7 Program memory Outlook Sparse read/write Variational memory Neural Turing Machine Agenda
- 8. Motivation: Dialog system 8/06/2019 8 A dialog system needs to maintain the history of chat (e.g., could be hours) Memory is needed The generation of response needs to be flexible, adapting to variation of moods, styles Current techniques are mostly based on LSTM, leading to “stiff” default responses (e.g., “I see”). There are many ways to express the same thought Variational generative methods are needed. vectorstock
- 9. Variational memory encoder- decoder (VMED) 8/06/2019 9 Conditional Variational Auto-Encoder contextgenerated latent variables VMED contextgenerated latent variables memory reads
- 11. 8/06/2019 11 Program memory Outlook Sparse read/write Variational memory Neural Turing Machine Agenda
- 12. Problems of current NTMs Lack of theoretical analysis on optimal memory operations. Previous works are based on intuitions: Location-based reading/writing; temporal linkage reading; least-used writing [Santoro et.al, Graves et.al] Sparse access over big memory [Rae et.al] Very slow due to heavy memory read/write computations 12
- 13. Cached Uniform Writing (CUW) 13
- 14. Ablation Study Memory-augmented Neural Networks w/wo Uniform Writing 14Task: repeat the input sequence twice
- 15. Synthetic tasks: memorize all 15
- 16. Synthetic tasks: memorize selectively 16
- 17. Synthetic sinusoidal generation: memorize featured points 17
- 20. 8/06/2019 20 Program memory Outlook Sparse read/write Variational memory Neural Turing Machine Agenda
- 21. Computing devices vs neural counterparts FSM (1943) ↔ RNNs (1982) PDA (1954) ↔ Stack RNN (1993) TM (1936) ↔ NTM (2014) UTM/VNA (1936/1945) ↔ NUTM--ours (2019) The missing piece: A memory to store programs Neural stored-program memory
- 22. NUTM = NTM + NSM
- 23. Multi-level modelling Hierarchical Regression: if the input is clustered, clustering before regression helps Prove for low dimensions maybe available, higher dimension?
- 24. NSM is beneficial to NTM
- 26. Sequencing tasks
- 29. Question answering (bAbI dataset)
- 30. 8/06/2019 30 Program memory Outlook Sparse read/write Variational memory Neural Turing Machine Agenda
- 31. Memory for graphs & relational structures Turing machine to design machine learning algorithms Memory-supported reasoning Imaginative memory Social memory: collective mem, theory of mind, memory of others Full cognitive architectures Theoretical analysis 8/06/2019 31 https://twitter.com/nvidia/status/1010545517405835264 Towards AGI: Is Human Brain a (super-)Turing machine?