All Questions

10,202 questions
Filter by
Sorted by
Tagged with
1 vote
18 views

26 views

Play against your own RL-trained AI from gym retro

so far I have seen people implementing reinforcement learning to build an AI to play and complete games on gym retro, such as street fighter, racing games and so on. However, I was wondering if it is ...
1 vote
17 views

what is Normalize for in PyTorch transfer learning tutorial?

in this pytorch tutorial, there is transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]), what is the purpose of this? (i removed it and the code still ...
11 views

why cross entropy loss has to be multiplied by a batch size during an evaluation in transformer model?

I am trying to look through a code of the transformer model from Pytorch. However, I do not understand why batch size needs to multiply with cross-entropy loss given that loss is calculated based on ...
• 1
9 views

How to propagate backwards in a neural network with an error term based on the average error over an episode of actions?

I am writing a neural network. I have an average error over an episode of actions to work with in order to update my weights. I know that in a 1 step neural network I take the most recent action of ...
• 3
29 views

Reinforcement learning SOTA with continuous action space

as of July 2022 what is the SOTA in reinforcement learning with continuous action space? DDPG PPO … other?
• 101
46 views

Reconstructing 3D models from 2D images using autoencoders

I went through a research paper ("Voxel-Based 3D Object Reconstruction from Single 2D Image Using Variational Autoencoders") and tried to implement the approach following this diagram: ![...
19 views

is it possible to train the same neural network with different numbers of inputs and outputs?

is it possible to create an adaptative neural network that can change the number of its inputs and outputs without having to train it each time it changes? the neural netwrok has to take purchases and ...
16 views

Markov decision process how to get the correct policy if targets are reached once among N episodes?

I have implemented an MDP on a network such that an agent starts in a node, takes an action from a set of predefined actions and next node (including current). Some of the nodes would result in ...
35 views

Good morning. I am a student running a project using medical data, predicting if the patient will or won't get a disease. The data has about 50k cases and 70 features. I proposed to train 5 models- ...
18 views

Is there benefit to autoregressive models for deep RL tasks with long episodes and short required context?

General Case In deep RL (specifically in the space of policy gradient methods) it seems very common that encoder-decoder models (either transformer or RNN-variant) are used in the policy/value ...
• 101
154 views

Remove already reached targets from the system to enable reaching other targets?

This may be a very fundamental question, but somehow I can't decide. I have a graph and the user can take several actions while traversing it and there are multiple points with rewards. When I execute ...
17 views

Effective fast mobile ocr model

I plan to develop OCR application which is mobile oriented and fast. (like 10~30 fps) The images that will be detected is not wild images. They are refined data such as cell phone capture images. In ...
287 views

How to deal with small reward values

In my environment rewards are generally small, e.g. [-0.01, 0.01]. My concern is that small reward values might get dominated or distorted by the noise during the training. Does it make sense to scale ...
• 197
1 vote
22 views

Rationalle behind SE3 Transformer?

I have just finished reading the SE3 transformer paper (SE(3)-Transformers: 3D Roto-Translation Equivariant Attention Networks) by Fuchs et-al and although I'm sure I understand less than 50% of the ...
• 121
13 views

Keras Multiclass Classification - More units than classes on last layer

I am building a CNN to classify spectograms and using the following architecture currently: ...
1 vote
27 views

Help with model architecture for a racing game

I’m working on a model for a racing game using pytorch. The model gets frame from the game as input and produces a controller state as output. The dataset consists of frames from the game and ...
• 11
17 views

Tracing the source: Which reference did the idea of Transformer's Key-query-value come from?

Since Transformers was proposed in 2017, there have been various interpretation schemes about KQV, but the original text does not seem to explain in detail what this KQV is inspired by. I don't need ...
• 101
4 views

Can kernel methods be used for prediction on unlabeled data?

In machine learning, kernel methods are often used in supervised learning, especially SVM. I would like to ask can kernel methods be used for prediction on unlabeled data under the premise of ...
• 101
7 views

How should classification on small images be done?

I want to create an image classifier that classifies very small images (16-32 pixels/side) into around 200 categories. Every category has exactly one image that defines it. The classifier should ...
• 101
16 views

What is a good neural network approach for this time-based data series

I’m trying to work out a neural network approach to a particular problem and would appreciate any advice. I have a machine that collects data over a period of time using 2 sensors. Data is collected ...
74 views

Can an AI have awareness

I guess my question can come under philosophy too. I was thinking about the following: According to Sir Roger Penrose "No computer has any awareness of what it does.". Now some context to ...
• 121
18 views

Different Kernel Initializers in my prediction layer with Transfer Learning could affect performance?

So I have this model right here and the task is to classify 3 labels.: ...
1 vote
9 views

Cannot understand/ reproduce reuslt in figure 1 from MobileNetV2?

Hi I am recently reading the MobileNetV2 paper and I found I cannot understand the following figure in the paper. I made a simple code trying to reproduce the results, but I got opposite results. ...
1 vote
12 views

Is the loss calculation step in Logistic Regression even needed?

I was reading about Logistic Regression and trying to implement the model from scratch. Maybe I am wrong, but I have noticed that the loss calculation step is meaningless in training a Logistic ...
12 views

Continuous-valued labels for supervised VAEs?

Referring to these two papers out of several similar ones: Ilse, Maximilian, et al. "Diva: Domain invariant variational autoencoders." Medical Imaging with Deep Learning. PMLR, 2020. http://...
1 vote
46 views

Time taken to solve cartpole environment using DQN

I am trying to solve the cartpole environment (GitHub) using DQN agent. I have been building my own DQN agent by following a tutorial by Jon Krohn. I am able to solve the environment with a maximum ...
13 views

Simplest approach to sentence inversion

I'm looking for an open source library/framework to perform (what I think is called?) sentence inversion. What I'm looking to do is take a sentence like: ...
• 101
12 views

Why episode mean rewards drawdown?

I'm new to RL. I'm using RecurrentPPO with parameter MlpLstmPolicy and the other defaults. Why the ...
• 123
24 views

Metrics using batches v/s metrics using full dataset

I am using training an image classification model using the pre-trained mobile network. During training, I am seeing very high values (more than 70%) for Accuracy, Precision, Recall, and F1-score on ...
• 151
1 vote
29 views

LSTM exploding? - multiple parallel time series with multiple variables

I have the following situation: Stock Time_Stamps Feature_1 Feature_2 Feature_n Price Stock_1 2019 0.5 1.0 1.0 100 Stock_1 2020 0.7 1.3 0.9 90 Stock_2 2019 0.3 0.9 1.1 110 Stock_2 2020 0.2 0.8 1....
25 views

Emergent behavior in AI models that looks similar to natural neural systems

"ImageNet Classification with Deep Convolutional Neural Networks" by Krizhevsky & Sutskever & Hinton describes very interesting emergent behavior of the AlexNet. It was trained on 2 ...
• 103
1 vote
83 views

What type of neural network has an unorganized structure?

I am looking for a network that has an unorganized structure like this, is feed-forward, does not have back-propagation functionality, and is trained with a genetic algorithm. What would I be looking ...
• 17
45 views

How special tokens in BERT-Transformers work?

"[SEP] tokens are useful to differentiate the questions from answers through type_ids" Yes, but how is this helping model to understand that "I should look paragraph and generate ...
• 1
1 vote
58 views

Can we combine Alpha-zero with GTP-4 to create a general AI?

Alpha Zero is good at looking into the future to plan it's next move. GTP-4 is good at generating language from previous text. It seems like combining these two systems would create a general ...
• 2,074
12 views

Tangent/slope at a point of a recurrent neural network

I am using a recurrent neural network for data of the form $\{(x_t, y_t)\}_{t=1}^T$. I defined the input of the RNN as a sequence $(x_{t-1}, x_t, y_{t-1})$ and output as $y_t$. My RNN has therefore ...
30 views

which method are able to approximate the following sequence 1 0 1 0 0 1 0 0 0 1 0 0 0 0 1 5x0 1 6x0

I'm learning deep learning and machine learning techniques, but it looks like neural networks are not capable of predicting the next element in the sequence ...
984 views

Deep Learning with Best-so-far instead of Where-you-are

It is my understanding that when training a Deep NN in Tensorflow/PyTorch/... we only keep the current state of the network in memory, except perhaps when we manually decide to save the current ...
• 172

15 30 50 per page