Questions tagged [convergence]

For questions related to the convergence of AI algorithms.

Filter by
Sorted by
Tagged with
2
votes
1answer
292 views

Why does Q-learning converge to the optimal policy, even if the agent acts sub-optimally?

In Q-learning, during training, it doesn't matter how the agent selects actions. The algorithm always converges to the optimal policy. Why does this happen? What's the intuition?
5
votes
2answers
180 views

Does the policy iteration convergence hold for finite-horizon MDP?

Most RL books (Sutton & Barto, Bertsekas, etc.) talk about policy iteration for infinite-horizon MDPs. Does the policy iteration convergence hold for finite-horizon MDP? If yes, how can we derive ...
3
votes
1answer
256 views

Why does the error of my LSTM not decrease after 10 epochs?

Despite the problem being very simple, I was wondering why an LSTM network was not able to converge to a decent solution. ...
2
votes
0answers
205 views

Why is it hard to prove the convergence of the deep Q-learning algorithm?

Why is it hard to prove the convergence of the DQN algorithm? We know that the tabular Q-learning algorithm converges to the optimal Q-values, and with a linear approximator convergence is proved. ...
4
votes
1answer
87 views

REINFORCE algorithm for portfolio optimization - problem while training

I'm trying to implement the Reinforce algorithm (Monte Carlo policy gradient) in order to optimize a portfolio of 94 stocks on a daily basis (I have suitable historical data to achieve this). The idea ...
4
votes
1answer
174 views

Why does Q-learning converge under 100% exploration rate?

I am working on this assignment where I made the agent learn state-action values (Q-values) with Q-learning and 100% exploration rate. The environment is the classic gridworld as shown in the ...
6
votes
2answers
5k views

What is convergence in machine learning?

I came across this answer on Quora, but it was pretty sparse. I'm looking for specific meanings in the context of machine learning, but also mathematical and economic notions of the term in general.
2
votes
1answer
65 views

Why do I get the best policy before Q values converge using DQN?

I have implemented DQN algorithm and wonder why during testing, the best performance is achieved by a policy from about 300 episode, when mean Q values converge at about 800 episode? Mean Q-values ...
0
votes
1answer
57 views

Is it a good idea to overfit on a small part of your data for faster model convergence?

I working on a classification problem that needs to detect patterns on a time serie. Basically, there's a catch-all class that means "no pattern detected", the other are for the specific patterns. The ...
2
votes
1answer
771 views

Deep Q-Learning poor convergence on Stochastic Environment

I'm trying to implement a Deep Q-network in Keras/TF that learns to play Minesweeper (our stochastic environment). I have noticed that the agent learns to play the game pretty well with both small and ...
1
vote
1answer
66 views

If $\alpha$ decreases over time, why is Q-learning guaranteed to converge?

Q-Learning is guaranteed to converge if $\alpha$ decreases over time. On page 161 of the RL book by Sutton and Barto, 2nd edition, section 8.1, they write that Dyna-Q is guaranteed to converge if each ...
7
votes
1answer
140 views

Why does reinforcement learning using a non-linear function approximator diverge when using strongly correlated data as input?

While reading the DQN paper, I found that randomly selecting and learning samples reduced divergence in RL using a non-linear function approximator (e.g a neural network). So, why does Reinforcement ...
3
votes
3answers
221 views

Why can the Bellman equation be turned into an update rule?

In chapter 4.1 of Sutton's book, the Bellman equation is turned into an update rule by simply changing the indices of it. How is it mathematically justified? I didn't quite get the initiation of why ...
1
vote
0answers
18 views

React on train-validation curve after trening

I have a regression task that I tray to solve with AI. I have around 6M rows with about 30 columns. (originally there was 100, but I reduce it with drop feature importance) I understand basic ...
1
vote
0answers
39 views

How to have closer validation loss and training loss in training a CNN

I am using an AlexNet architecture as my Convolutional Neural Network. A learning rate of 0.00007 and 128 batch_size. I have 20000 data and 10% test, 40% validation, and 50% for training. I used 100 ...
3
votes
1answer
244 views

How to determine if Q-learning has converged in practice?

I am using Q-learning and SARSA to solve a problem. The agent learns to go from the start to the goal without falling in the holes. At each state, I can choose the action corresponding to the maximum ...
1
vote
0answers
49 views

Is Monte Carlo tree search guaranteed to converge to the optimal solution in two player zero-sum stochastic games?

I'm aware that convergence proofs for Monte Carlo tree search exist in the case of deterministic zero sum games and Markov decision processes. I have come across research which applies MCTS to zero-...
2
votes
1answer
29 views

How much can an inclusion of the number of iterations have on the training of an MLP?

My doubt is like this : Suppose we have an MLP. In an MLP, as per the backprop algorithm (back-propagation algorithm), the correction applied to each weight is : $$ w_{ij} := -\eta\frac{\partial E}{\...
4
votes
1answer
228 views

When do SARSA and Q-Learning converge to optimal Q values?

Here's another interesting multiple-choice question that puzzles me a bit. In tabular MDPs, if using a decision policy that visits all states an infinite number of times, and in each state, randomly ...
1
vote
0answers
32 views

How can deep Q-learning converge if the targets may not be correct?

In deep Q-learning, $Q(s, a)$ and $Q'(s, a)$ are predicted or estimated by the neural network itself. In supervised learning, the target value is a true unbiased value. However, this isn't the case in ...
2
votes
0answers
352 views

Why is DDPG not learning and it does not converge?

I have used a different setting, but DDPG is not learning and it does not converge. I have used these codes 1,2, and 3 and I used different optimizers, activation functions, and learning rate but ...
2
votes
0answers
56 views

If the performance of an RL agent in a partially observable environment is “good”, is this likely only accidental?

In my research, I remember to have read that, in case of an environment which can be modeled by partially observable MDP, there are no convergence guarantees (unfortunately, I do not find the paper ...
2
votes
1answer
605 views

Is there a simple proof of the convergence of TD(0)?

Does anybody know a simple proof of the convergence of the TD(0) value function prediction algorithm?
6
votes
0answers
161 views

Is the Bellman equation that uses sampling weighted by the Q values (instead of max) a contraction?

It is proved that the Bellman update is a contraction (1). Here is the Bellman update that is used for Q-Learning: $$Q_{t+1}(s, a) = Q_{t}(s, a) + \alpha*(r(s, a, s') + \gamma \max_{a^*} (Q_{t}(s', ...
2
votes
1answer
46 views

What is convergence analysis, and why is it needed in reinforcement learning?

While reading a paper about Q-learning in network energy consumption, I came across the section on convergence analysis. Does anyone know what convergence analysis is, and why is convergence analysis ...
2
votes
0answers
84 views

If the minimum Q value is decreasing and the maximum Q value increasing, is this a sign that dueling double DQN is diverging?

I'm training a dueling double DQN agent with prioritized replay buffer and notice that the min Q values are decreasing, while the max Q values are increasing. Is this a sign that it is diverging? ...
1
vote
1answer
71 views

If deep Q-learning starts to choose only one action, is this a sign that the algorithm diverged?

I'm working on a deep q-learning model in an infinite horizon problem, with a continous state space and 3 possible actions. I'm using a neural network to approximate the action-value function. ...
5
votes
1answer
324 views

What are the conditions of convergence of temporal-difference learning?

In reinforcement learning, temporal difference seem to update the value function in each new iteration of experience absorbed from the environment. What would be the conditions for temporal-...
3
votes
0answers
61 views

Convergence of a delayed policy update Q-learning

I thought about an algorithm that twists the standard Q-learning slightly, but I am not sure whether convergence to the optimal Q-value could be guaranteed. The algorithm starts with an initial ...
2
votes
1answer
534 views

Why isn't my implementation of A2C for the the atari pong game converging?

I have two different implementations with PyTorch of the Atari Pong game using A2C algorithm. Both implementations are similar, but some portion are different. https://colab.research.google.com/drive/...
4
votes
1answer
201 views

When exactly is a model considered over-parameterized?

When exactly is a model considered over-parameterized? There are some recent researches in Deep Learning about the role of over-parameterization toward generalization, so it would be nice if I can ...
3
votes
1answer
149 views

Why do RL implementations converge on one action?

I have seen this happening in implementations of state-of-the-art RL algorithms where the model converges to a single action over time after multiple training iterations. Are there some general ...
1
vote
0answers
58 views

When does Monte Carlo linear function approximation converge?

In this Stanford lecture (minute 35:47 and 37:00), the professor says that Monte Carlo (MC) linear function approximation does not always converge, and she gives an example. In general, when does MC ...
1
vote
1answer
96 views

Why isn't the implementation of my policy evaluation for a simple MDP converging?

I am trying to code out a policy evaluation algorithm to find the $V^\pi(s)$ for all states. The following diagram below shows the MDP. In this case i let p = q = 0.5. the rewards for each states ...
3
votes
1answer
151 views

Neural network doesn't seem to converge with ReLU but it does with Sigmoid?

I'm not really sure if this is the sort of question to ask on here, since it is less of a general question about AI and more about the coding of it, however I thought it wouldn't fit on stack overflow....
2
votes
1answer
265 views

Is there an advantage in decaying $\epsilon$ during Q-Learning?

If the agent is following an $\epsilon$-greedy policy derived from Q, is there any advantage to decaying $\epsilon$ even though $\epsilon$ decay is not required for convergence ?
1
vote
0answers
66 views

Does SARSA(0) converge to the optimal policy in expectation if the Robbins-Monro conditions are removed?

The conditions of convergence of SARSA(0) to the optimal policy are : The Robbins-Monro conditions above hold for $α_t$. Every state-action pair is visited infinitely often The policy is greedy with ...
1
vote
2answers
166 views

What are the conditions for the convergence of SARSA to the optimal value function?

Is it correct that for SARSA to converge to the optimal value function (and policy) The learning rate parameter $\alpha$ must satisfy the conditions: $$\sum \alpha_{n^k(s,a)} =\infty \quad \text{and}...
0
votes
1answer
34 views

Is it possible to use deeplearning with spark (with a distributed databases as HDFS or Cassandra)?

If it is possible, will it be really useful or the model will end up converging very early(with a typical optimum learning rate) ? Any content on this topic will be helpful for me.
1
vote
1answer
139 views

Does TD(0) prediction require Robbins-Monro conditions to converge to the value function?

Does the learning rate parameter $\alpha$ require the Robbins-Monro conditions below for the TD(0) algorithm to converge to the true value function of a policy? $$\sum \alpha_t =\infty \quad \text{...
4
votes
1answer
171 views

Convergence of semi-gradient TD(0) with non-linear function approximation

I am looking for a result that shows the convergence of semi-gradient TD(0) algorithm with non-linear function approximation for on-policy prediction. Specifically, the update equation is given by (...
1
vote
0answers
29 views

Imposing contraints on sequence of image classifications

Are there example implementations of networks that apply constraints across sequences of image classifications where class labels are ordinal numbers? For example, to cause the output of a CNN to ...
5
votes
1answer
49 views

How can we conclude that an optimization algorithm is better than another one

When we test a new optimization algorithm, what the process that we need to do?For example, do we need to run the algorithm several times, and pick a best performance,i.e., in terms of accuracy, f1 ...
0
votes
2answers
576 views

LSTM network doesn't converge, what should be changed? [closed]

I'm testing out TensorFlow LSTM layer text generation task, not classification task; but something is wrong with my code, it doesn't converge. What changes should be done? Source code: ...
1
vote
1answer
668 views

How is the actor-critic algorithm guaranteed to converge?

From my understanding, the critic evaluates the policy (actor) following dynamic programming (DP) or approximate dynamic programming (ADP) scheme, which should converge to the optimal value function ...
2
votes
1answer
155 views

Is there a rigorous proof for finding Hopfield minima?

I am looking for a rigorous mathematical proof for finding the several local minima of the Hopfield networks. I am searching for something rigorous, a demonstration, not just let the network keep ...
5
votes
1answer
513 views

How to show temporal difference methods converge to MLE?

In chapter 6 of Sutton and Barto (p. 128), they claim temporal difference converges to the maximum likelihood estimate (MLE). How can this be shown formally?
2
votes
0answers
43 views

How to show Monte Carlo methods converge to an estimate which minimizes mean squared error?

In chapter six of Sutton and Barto (p.128), they claim Monte Carlo methods converge to an estimate minimizing the mean squared error. How can this be shown formally? Bump
3
votes
0answers
30 views

How do we give a kick start to the Facenet network?

I read the Facenet paper and one thing I am not sure about (it might be trivial and I missed it) is how do we give the kick start to the network. The embeddings in the beginning are random, so ...
3
votes
0answers
37 views

How to know when a Environment will yield a deterministic model

Given enough experiment data on time taken for objects to fall to earth from different heights, one can create various models that will accurately predict the time it will take for an object falling ...