Optimal Tuning of Continual Online Exploration in Reinforcement Learning Youssef Achbany, Francois...
-
date post
20-Dec-2015 -
Category
Documents
-
view
220 -
download
0
Transcript of Optimal Tuning of Continual Online Exploration in Reinforcement Learning Youssef Achbany, Francois...
Optimal Tuning of Continual Online Exploration in
Reinforcement Learning
Youssef Achbany, Francois Fouss, Luh Yen, Alain Pirotte & Marco Saerens
Information Systems Research Unit (ISYS)Université de Louvain
Belgium
Achbany Youssef - UCL 2
Outline
Introduction Mathematical concepts Modelling exploration by entropy Optimal policy Preliminary experiments Conclusion and further work
Achbany Youssef - UCL 3
Introduction One of the challenges of reinforcement
learning is to manage: The tradeoff between exploration and
exploitation. Exploitation
aims to capitalize on already well-established solutions.
Exploration: aims to continually try new ways of solving the
problem. is relevant when the environment is changing.
Achbany Youssef - UCL 4
Introduction Simple routing problem
The goal is to reach a destination node (13) From an initial node (1) To minimize costs
For each node Set of admissible actions Weight (cost) associated We define a probability distribution on the set of admissible actions
1
2
5
4
3
9
8
7
6
10
11
12
14
13
1
11
1
1
11
1
1
22
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
Achbany Youssef - UCL 5
Mathematical concepts
We have a set of states, S = {1, 2, …,n} st = k means that the system is in state k
at time t
In each state s = k, we have a set of admissible control actions, U(k) So that u(k) U(k) is a control action
available at state k
Achbany Youssef - UCL 6
Mathematical concepts
When we choose action u(st) at state st, A bounded cost C(u(st)| st) < ∞ is incurred The system jumps to state st+1 = f(u(st)| st)
Where f is a function
We suppose the network of states does not contain any negative cycle
Achbany Youssef - UCL 7
Mathematical concepts
For each state s, we define a probability distribution on the set of admissible actions, P(u(s)| s)
Meaning that the choice is randomized This introduces exploration – not only
exploitation This is the main contribution of our
work
Achbany Youssef - UCL 8
Mathematical concepts
For instance if, in state s = k, there are three admissible actions,
The probability distribution P(u(k)| s=k) involves three values
k
uk1
P(uk
1|k)
uk2
uk3
P(u k3 |k)
P(uk2|k)
Achbany Youssef - UCL 9
Mathematical concepts
The policy is defined as the set of all probability distributions for all states
1
2
5
4
3
9
8
7
6
10
11
12
14
13
1
11
1
1
11
1
1
22
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
Achbany Youssef - UCL 10
Mathematical concepts The goal is to reach a destination state,
s = d From an initial state, s0 = k0
While minimizing the total expected cost
The expectation is taken on the policy, that is, on all the random variables u(k) associated to the states
Achbany Youssef - UCL 11
Mathematical concepts In other words, we have to determine
the best policy that minimizes V(k0) That is, the best probability distributions
This is standard, except the fact that we introduce choice randomisation
Achbany Youssef - UCL 12
Mathematical concepts We now introduce a way to control
exploration
We introduce the degree of exploration, Ek, defined on each state k Which is the entropy of the
probability distribution of actions in this state k
Achbany Youssef - UCL 13
Modelling exploration by entropy The degree of exploration, Ek, is
defined as the entropy at state k
The minimum is 0 (no exploration) The maximum is log(nk) where nk is the
number of admissible actions in state k (full exploration)
Achbany Youssef - UCL 14
Modelling exploration by entropy
While the exploration rate is defined as
and takes its value between 0 (no exploration)
and 1 (full exploration).
Achbany Youssef - UCL 15
Modelling exploration by entropy The goal now is to determine the
optimal policy under exploration constraints That is, seek the policy, *, among
for which the expected cost, V(k0), is minimal
while guarantying a given degree of exploration (entropy) in each state k
Achbany Youssef - UCL 16
Modelling exploration by entropy In other words,
where the Ek are provided/fixed by the user/designer
They control the degree of exploration at each node k
Achbany Youssef - UCL 17
Modelling exploration by entropy Thus, we route the agents as fast
as possible, while exploring the network
1
2
5
4
3
9
8
7
6
10
11
12
14
13
1
11
1
1
11
1
1
22
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
Achbany Youssef - UCL 18
Optimal policy Here are the necessary optimality
conditions (for a local minimum), very similar to Bellman’s equations V
*(k) is the optimal expected cost from state k
P(i|k) is the probability of chosing action i satisfying the entropy constraint through k
Achbany Youssef - UCL 19
Optimal policy
Which lead to the following updating rules Convergence has been proved in a
stationary environment
Achbany Youssef - UCL 20
Optimal policy This updating rule has a nice
interpretation: Route the agents preferably (with probability
P(i|k)) to the state from which the expected cost is minimal
Including the direct cost for reaching this state
1
2
5
4
3
9
8
7
6
10
11
12
14
13
1
11
1
1
11
1
1
22
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
Achbany Youssef - UCL 21
Optimal policy
If k is large (zero entropy: no exploration), we obtain
which is the common value iteration algorithm or Bellman’s equation
for finding the shortest path
Achbany Youssef - UCL 22
Optimal policy If k is zero (maximum entropy: full
exploration), We perform a blind exploration
We estimate the « average first passage time »
Without taking the costs into consideration:
where nk is the number of admissible actions in state k
Achbany Youssef - UCL 23
Advantages of our algorithm Our strategy could be interesting if the
environment is changing And there is a need for continuous exploration
Indeed, if no exploration is performed, The agent will not notice the changes unless
they occur on the shortest path So that the policy will not be adjusted
In other words, we propose an optimal exploration/exploitation trade-off
Achbany Youssef - UCL 24
Preliminary experiments Simple Network
routing Dynamic Uncertain
1
2
5
4
3
9
8
7
6
10
11
12
14
13
1
11
1
1
11
1
1
22
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
Achbany Youssef - UCL 25
Preliminary experiments Exploration rate of 0% for all nodes (no
exploration)
1
4
3
2
5
6
7
8
9
12
11
10
13
14
White: very low trafficLight gray: low trafficGray: medium trafficDark gray: high trafficBlack: very high traffic
Achbany Youssef - UCL 26
Preliminary experiments Entropy rate of 30% for all nodes
1
4
3
2
5
6
7
8
9
12
11
10
13
14
White: very low trafficLight gray: low trafficGray: medium trafficDark gray: high trafficBlack: very high traffic
Achbany Youssef - UCL 27
Preliminary experiments Entropy rate of 60% for all nodes
1
4
3
2
5
6
7
8
9
12
11
10
13
14
White: very low trafficLight gray: low trafficGray: medium trafficDark gray: high trafficBlack: very high traffic
Achbany Youssef - UCL 28
Preliminary experiments Entropy rate of 90% for all nodes
1
4
3
2
5
6
7
8
9
12
11
10
13
14
White: very low trafficLight gray: low trafficGray: medium trafficDark gray: high trafficBlack: very high traffic
Achbany Youssef - UCL 29
Preliminary experiments
Other experimental simulations are provided in: Tuning continual exploration in
reinforcement learning (Technical report submitted for publication).
http://www.isys.ucl.ac.be/staff/francois/Articles/Achbany2005a.pdf
Achbany Youssef - UCL 30
Conclusion In this work,
we presented a model integrating both exploration and exploitation in a common framework.
The exploration rate is controlled by the entropy of the choice probability distribution defined on the states of the system.
When no exploration is performed (zero entropy on each node), the model reduces to the common value iteration algorithm computing the minimum cost policy.
On the other hand, when full exploration is performed (maximum entropy on each node), the model reduces to a "blind" exploration, without considering the costs.
Achbany Youssef - UCL 31
Further work
This model has been extended to Stochastic shortest paths problems Discounted problems Acyclic graphs Edit-distances between string Developing links with Q-learning
Achbany Youssef - UCL 32
Thank you !!!