4

I am wondering how to generate datasets when the environment is not as simple as a tic-tac-toe or a maze problem There is no difference in concept, which is why tic-tac-toe and maze problems are used to teach. As you have noted, the main difference between reinforcement learning (RL) and supervised learning is that RL does not use labeled datasets. If you ...


4

You can already do this with some neural networks, such as GANs and VAEs, which are generative models that learn a probability distribution over the inputs, so they learn how to produce e.g. images that are similar to the images they were trained with. Now, if you're interested in whether there is a black-box method, i.e. a method that, for every possible ...


4

They are equivalent. When we consider a particular instance as a vector, we are not literally imagining it as an arrow with it's head at the point coordinates and tail at the origin. It's just when you are working with a tuple of numbers in a mathematical context, it is conventional to call it a vector. This language follows into machine learning which is ...


3

It's true that your original dataset can contain duplicates, so it should not be called a set, in order to be consistent with the mathematical definition of a set. There are mathematical objects known as multi-sets that can contain duplicates, but the order of the elements is still not relevant. There are also tuples and sequences, where the order of the ...


2

There are quite a few examples of papers where they try and 'teach' neural networks to 'learn' how to solve math problems. Most of the time, sadly, it comes down to training on a large dataset after which the network can 'solve' the sort of basic problems, but is unable to generalize this to larger problems. That is, if you train a neural network to solve ...


2

@nbro pointed out the paper A systematic study of the class imbalance problem in convolutional neural networks, which tested class imbalance LeNet for MNIST, on a custom CNN for CIFAR-10, and on ResNet for ImageNet. The paper found that by artificially creating class imbalance on those data sets, the neural networks are significantly deteriorated. The ROC ...


2

Short answer Shuffling affects learning (i.e. the updates of the parameters of the model), but, during testing or validation, you are not learning. So, it should not make any difference whether you shuffle or not the test or validation data (unless you are computing some metric that depends on the order of the samples), given that you will not be computing ...


2

If I understood correctly, the model is a polynomial equation No, it's not true that all machine learning (ML) models compute (or represent) a polynomial function. For example, a sigmoid is not a polynomial, but, for example, in a neural network, you can combine many sigmoids to build complicated functions that may not necessarily be polynomials. We usually ...


2

The paper Evolutionary Dataset Optimisation: learning algorithm quality through evolution (2019), by Henry Wilde et al., proposes a method to generate datasets with a genetic algorithm. Their goal is to generate data for which a particular algorithm performs well, in terms of a certain metric, so that to get more insights about this algorithm and why it ...


2

This question raises a lot more questions. It seems like a solution looking for a problem, instead of the other way round. How do you measure the fitness of a feature? What would one of the "evolved datasets" mean? What does it represent? What would your overal purpose be? If you just wish to generate simulated datasets, there are easier ways to ...


2

How can I generate the target label from the other data in the dataset? If you are asking how you can create the learning signal in SSL, when given an unlabelled dataset, for learning representations of these unlabelled data, then there is no general answer. The answer depends on the type of data that you have (which can be e.g. textual or visual), and ...


2

Unfortunately, the answer here is that "it depends". People have taken different approaches to this problem and I'll describe a few here. None of which however is the "right" answer. Labeling When generating benchmark datasets, we actually do have this problem. To be honest, most of the time the labeling is done to the best ability of the ...


2

You can see some labels at https://www.tensorflow.org/datasets/catalog/emnist. It goes like this: ‘0’-‘9’ are 0-9 ‘A’-‘Z’ are 10-35 ‘a’-‘z’ are 36-61


2

You can claim to use a real-world dataset, you would just need to specify that some values were interpolated. Do you have to have the inter-mediate values though? By the looks of it, each "region" was only measured every 2 hours, so I would just keep it that way and just have the resolution be 2 hours. It doesn't have to be hourly, and probably ...


2

You have implicitly assumed that supervised learning is being used, given the assumption that labels are needed. But this might lead to the following potential problems: Log file data tends to be huge, and it may be infeasible to label due to the time/expertise required; Then there's the class imbalance problem, in that attack examples are far far rarer ...


2

Can residual connections be beneficial when we have a small training dataset? The usual rule of data science investigations applies here: Try it, measure the results, then you will know. It is very hard to tell, a priori, whether a specific architectural or hyperparameter choice will impact the performance of a neural network on a given problem. In this ...


2

The number 50 is essentially just a guess based on results when compressing and/or generating data of a certain type. The variables such as "the three translations of the body, the three rotations of the head and the independent movements of the face's muscles" are examples only. There is no known formal map with well-defined parameters that ...


1

The good folks behind Spacy have their paid product called Prodigy which is a data labeling tool. I haven't used it but it appears you can host it somewhere and then you would just have to send the link to the students. It is a little pricey but you get a lifetime license... A free alternative might be Label Studio but I am not sure how easy it is to host it ...


1

I personally can’t think of any good reason to apply data augmentation before splitting the dataset, though one may exist. The issue is that if you augment first and split later, you risk introducing unwanted correlations between your training and test datasets. In the paper you linked it sounds as if the training data is all procedurally derived from the ...


1

In my personal experience, that depends. Augmenting data for training purposes is valid, and can even improve performance, as you may be aware. For testing purposes, it may be valid. Let me give you two examples when that may be the case: Facial Recognition. Imagine that you have an augmentation function that can change the face pose (left/right pose, for ...


1

I know at least one example where the rank of the dataset (more specifically, the rank of a matrix that is computed from the design matrix, i.e. the matrix with your data, which I will describe more in detail below) can have an impact on the number of solutions that you can have or how you find those solutions. I am thinking of linear regression. So, in ...


1

As for me, the easiest path to what you are asking for is generating them yourself. An Example I usually grab some TTF fonts and put them into a directory so that I have variety for character identification. Begin by importing dependencies and creating a generator function: from PIL import Image, ImageDraw, ImageFont from os import listdir WIDTH = ...


1

DALL·E is a 12-billion parameter version of GPT-3 trained to generate images from text descriptions should be the same data they used to train the GPT-3


1

There are two possible forms of overfitting. First related to the training only (fitting weights) and second related to architecture (fitting hyperparameters) and these two must be checked in two different stages. When you check performance of the given model you have two do this on unseen data, so you fit weights on training data and check it on validation ...


1

Usually we are splinting the training dataset because we want to fine tune, find the best hyper-parameters, for our model. If we combine the validation and training dataset given a network complex enough we could achieve perfect performance for the given task. But having very good performance on the training dataset does not mean our model is useful. It ...


1

I'm not aware of a direct way for finding the best NN architecture for a given task, but the recommended way, as far as I know, is to devise a network that can overfit the training data, and then apply regularization on top of it. That way, you can be almost sure you're not underfitting/underperforming due to network capacity.


1

The paper referenced by Martin Thoma is the go-to for semantic segmentation. However I will also like to add the Panoptic Segmentation metric as an aggregated method to measure both the detection task and segmentation task of the model. It is a very well-known and widely used metric since it is the standard metric for COCO dataset (segmentation) This is the ...


Only top voted, non community-wiki answers of a minimum length are eligible