Since the early days of Artificial Intelligence, designing an agent able to interact with humans was considered a remarkable goal, aiming to replicate human cognitive capabilities in computer. In this work, we outline the main topics of a prospective PhD project whose aim is to exploit Deep Reinforcement Learning techniques to learn to interact with users in Natural Language conversations.
LANGUAGE ASSIMILATION THROUGH TYPE-2 FUZZY GRAMMARS: AN APPLICATION IN COMPUT...paperpublications3
Similar to Empowering Conversational Agents with Situated Natural Language Communication Skills by Exploiting Deep Reinforcement Learning Techniques (20)
Empowering Conversational Agents with Situated Natural Language Communication Skills by Exploiting Deep Reinforcement Learning Techniques
1. Empowering Conversational Agents with Situated Natural Language
Communication Skills by Exploiting Deep Reinforcement Learning
Techniques
Alessandro Suglia
Heriot-Watt University, Edinburgh Centre for
Robotics, Edinburgh, Scotland, UK
D o c t o ra l C o n s o r t i u m A I * I A 2 0 1 7 , B a r i
Supervisor: Prof. Oliver Lemon
Director of the Interaction Lab,
Heriot-Watt University
2. Dynamic Temporal Contextualized
The answer provided
by one of the
interlocutor affects
the state of the
dialogue
What is dialogue?
The interpretation of
each utterance
depends incredibly on
different environment
conditions
Different topics are
discussed in a given
conversation without
a specific order
[1]: Rieser, V., and O. Lemon. Reinforcement learning for adaptive dialogue systems: a data-driven methodology
for dialogue management and natural language generation. Springer Science & Business Media, 2011.
3. Goal-directed learning from interaction with an
environment by learning to map situations to actions
by exploiting Deep Neural Networks models
Deep Reinforcement Learning
4. Deep Reinforcement Learning
Successfully applied to different games such as Go [2] or Poker [3]
Incrementally abstract representations learned from training data
Learning to behave by interacting with the environment
Learning in an incremental and online fashion
Inability to generalise due to the domain specific design
Handsome quantity of data required to effectively learn a policy
[2]: Silver, D., Huang, A., Maddison, C.J., Guez, A., Sifre, L., Van Den Driessche, G., Schrittwieser, J., Antonoglou, I.,
Panneershelvam, V., Lanctot, M., et al.: Mastering the game of Go with deep neural networks and tree search. Nature
529(7587) (2016) 484–489
[3]: Moravčík, Matej, et al. "Deepstack: Expert-level artificial intelligence in no-limit poker." arXiv preprint arXiv:
1701.01724 (2017).
5. RQ1 : Is it possible to train a conversational agent to interact
within real world contexts consisting of embodied agents and
situated objects?
6. RQ2: Is it possible to train a conversational agent to generate
accurate contextualized responses for the user?
7. RQ3: Is the system able to adapt seemly to different domains by
exploiting what it has previously learned?
9. Thanks!
You can find me at the poster session
@ale_suglia
as247@hw.ac.uk
Any questions?
10. References
[1]: Rieser, V., and O. Lemon. Reinforcement learning for adaptive dialogue systems: a data-driven
methodology for dialogue management and natural language generation. Springer Science &
Business Media, 2011.
[2]: Silver, D., Huang, A., Maddison, C.J., Guez, A., Sifre, L., Van Den Driessche, G., Schrittwieser, J.,
Antonoglou, I., Panneershelvam, V., Lanctot, M., et al.: Mastering the game of Go with deep neural
networks and tree search. Nature 529(7587) (2016) 484–489
[3]: Moravčík, Matej, et al. "Deepstack: Expert-level artificial intelligence in no-limit poker." arXiv
preprint arXiv:1701.01724 (2017).