Questions tagged [training]
Training is the part of machine learning whereby a model is "trained" on a define portion of a dataset to learn attributes and statistical features of the data. It's counterparts are called Testing and Validation. After training a model is tested and validated on another portion of the dataset.
691 questions
0 votes
0 answers
58 views
Why bias value is critical to successful learning?
Given basic elements of a neuron(as below) with a bias value: I learnt that, a bias value allows you to shift the activation function(say sigmoid function) to the left or right, which may be critical ...
- 199
3 votes
0 answers
37 views
Why is there no spatial batch layer norm?
We already have (spatial) batch norm and (spatial) layer norm: Why don't we normalize over everything so that each entire activation plane over all batches over all channels gets the benefits of both ...
- 31
2 votes
0 answers
34 views
KDE classification with n>1 features
i'm working on an implementation of this paper and i have a question. The authors purpose a model (KDE boosting classifier) which works with only n=1 feature and 1 dependent variabile. I'm saying that ...
- 21
-2 votes
1 answer
106 views
how to tell whether artificial neural network is suitable for a particular problem?
Usecase - Gmail detecting spam format emails is a simple NLP system based on some clear rules. Doesn't rely on artifical neural network. No training involved. Usecase - Gmail predicting auto-...
- 199
4 votes
1 answer
632 views
Why does the function of a neuron need to be in range [-1,1] or [0,1]?
The concept of a function includes three major components: a set called “domain”, a set called "codomain" and a rule which for each element of a domain points to (puts into a correspondence)...
- 199
2 votes
1 answer
66 views
How to determine the optimal number of training epochs when validation loss stabilizes but does not increase?
I’m training a CNN (DenseNet169) for a medical imaging task with ~12,000 training samples using fine-tuning (pretrained on ImageNet). I monitor both training and validation loss/accuracy. What I see ...
- 331
7 votes
1 answer
103 views
Effects of resizing training images during preprocessing CNN classification model
I'm trying to train a CNN model to identify phytoplankton species from a training set. During preprocessing, the images are resized to 224x224, which seems to be stretching or compressing the object ...
- 115
3 votes
1 answer
268 views
Why does one of the fold in my k-fold cross validation consistently train longer?
I'm working on a binary classification problem using LightGBM with 5-fold cross-validation. My dataset is highly imbalanced, with approximately 1,000 positive samples and 375,000 negative samples. ...
- 101
4 votes
1 answer
59 views
When same combination of variable values appear in train and test datasets?
I’m studying the basics of ML and trying to train a random forest model in a .csv dataset which each row contains the values of pixels in the red, green and blue bands (all varying from 0-255 values) ...
- 43
11 votes
4 answers
2k views
Does training a neural network on a combined dataset outperform sequential training on individual datasets?
I have a neural network with a fixed architecture (let's call it Architecture A). I also have two datasets, Dataset 1 and Dataset 2, both of which are independently and identically distributed (i.i.d.)...
0 votes
0 answers
29 views
Stale weights and gradients given Adam with an optimal learning rate
I'm fitting a network to predict a delta between eight corresponding 3D points at two timesteps. The model consists of two MLPs with two layers each, with LeakyRELU in between the layers. It takes in ...
- 61
2 votes
0 answers
51 views
Reducing loss of multiclass classication problem
I have been reading a paper in which they theoretically showed existence of a Neural network model that can perform a algorithm which involves selection of stencils with 100% accuracy. So they ...
- 31
6 votes
1 answer
110 views
Making sense of a flat validation curve given improving training
The figures below depict validation and training curves for metrics (top row, the lower the better) and losses (bottom row). The last column depicts aggregated metrics/losses from the first two ...
- 61
0 votes
0 answers
36 views
Validation metrics plateau from the first few epochs at relatively good values and don't improve
I am working on 6D pose tracking, where the goal is to estimate how 3D position and orientation of an object changes from frame t-1 to t. Train/validation datasets are synthetic and come from a single ...
- 61
0 votes
0 answers
65 views
Logistic Regression Loss can be zero? Question from a Test
I have a question from a test, I managed to solve it, but something feels weird... Prove it is false: If all the samples for Logistic Regression are categorized false, so the training loss is 0. What ...
