Questions tagged [deep-learning]

a new area of Machine Learning research concerned with the technologies used for learning hierarchical representations of data, mainly done with deep neural networks (i.e. networks with two or more hidden layers), but also with some sort of Probabilistic Graphical Models.

Filter by
Sorted by
Tagged with
0 votes
0 answers
11 views

Predicted images are quite good with loss=0.20 while are black with loss=0.02

I'm trying to train a U-net with VGG16 as a backbone in order to recognize 4 classes: sky, rocks, trees and background in a dataset of about 10000 images. I'm using categorical crossentropy as a loss ...
  • 111
1 vote
1 answer
24 views

Combine datasets of different domains to ehance generalizibility

so I try to implement an Emotion Classifier, which should detect several emotions from a text. There are several datasets for this (ISear, GoEmotions, etc.). However, a lot of them come from different ...
  • 13
0 votes
0 answers
20 views

Why am I seeing these spikes in model loss curve

I am training an image classifier on 1152 images in 4 classes , I have used data augmentation too . ...
0 votes
0 answers
21 views

How can create deliberately biased models?

I deal with an image classification problem with 3-class. I want to create a model which takes side to one specific class. I mean, while the model predicts a sample, if it is hesitant between class-1 ...
  • 101
3 votes
3 answers
438 views

How to train deep learning model on high dimensional dataset with limited memory and disk

For large datasets in terms of rows, usually it is handled by splitting data into pieces and feeding them into the model one at a time using tf.datasets or custom generator. However what if number of ...
  • 385
0 votes
0 answers
36 views

Adaptive stopping algorithm find bound that holds with probability

The technique sets aside a validation set Sval, which is used to monitor the improvement of the training process. Let $h_1,h_2,h_3,...$ be a sequence of models obtained after $1,2,3,...$ epochs of ...
  • 101
-1 votes
0 answers
9 views

More dense layers with heavy dropouts or fewer layers with light dropouts?

I'm trying to build a network. While creating the fully connected part in the last, Which one should we prefer: More layers that regularly reduce with heavy dropouts or fewer layers that reduce ...
0 votes
1 answer
20 views

Difference between Siamese Network and Prototypical Networks for One Shot Learning

I am having a bit of trouble understanding how the architecture of prototypical networks in a one shot learning use case differs from Siamese networks. If I’m understanding correctly, Siamese networks ...
0 votes
1 answer
26 views

Increasing/Decreasing importance of feature/thing in ML/DL

I have 3 cases: I have a classification model that will be used to classify cats and dogs. On my train data dog pictures has a watermark on them, but cat pictures don't. The problem is: Whenever I ...
  • 89
2 votes
0 answers
17 views

Discussion about modern deep learning training strategies

Previously I have put a lot of effort into training networks appropriately. However, talking to colleagues, a lot of the things I did may be redundant due to novel optimizers and the theory of deep ...
0 votes
0 answers
7 views

Optimization of a dispatch strategy

I am working at a project where I have to optimize the makespan of a system. The briefly description of the system: -I have an Input Output Machine which sends tubes to analyzers which has to execute ...
0 votes
1 answer
20 views

Bounding box regression without a classification task

I am using PyTorch to create a model that detects certain objects in an image. I framed my problem as a regression on bounding boxes, without any classification task whatsoever. The reasoning behind ...
0 votes
1 answer
38 views

How do transformers differ from feature selection and regular machine learning?

This is perhaps a simplistic way of thinking, but to me transformers (attention based neural networks) focus on a subset of the input, learning what is important for the problem/prediction as the ...
  • 71
0 votes
0 answers
10 views

Why does the permutation symmetry breaking in dropout have a regularizing effect?

I usually include dropout in the dense layers I include in neural networks. I've taken it as conventional wisdom that this is a regularization. I've also compared model test performance on the same ...
  • 162
0 votes
0 answers
8 views

The val_accuracy is higher than training accuracy, and the test accuracy is very low compared to both val_accuracy and train_accuracy

I am training a CNN model where, Training data=687 , validation data=102 , testing data=79 The validation accuracy is higher than training accuracy The test accuracy is very low compared to both ...
0 votes
1 answer
49 views

One word changes everything NLP

I have a classification model (BERT) that classifies sentences as either question or normal sentences. But whenever a sentence has "how" word, the model chooses "question" class. ...
  • 89
0 votes
0 answers
15 views

Is it possible to "link-couple-connect" certain inputs with outputs in a MIMO seq2seq LSTM model?

I have a seq2seq model with encoder and decoder as LSTMs which takes INPUT as the past 4 days of building data (weather data, 5 zones data like occupancy, internal loads, indoor air temperatures, and ...
2 votes
1 answer
29 views

Question about non linearity of activation function

I have a basic question about activation functions. It is told that they are added to the network to introduce non linearity. However, the neural network itself is non linear. Isn' it? If we see any ...
-1 votes
1 answer
36 views

Sklearn vs Pytorch vs Tensorflow vs Keras

I just need to understand the differences between sklearn, pytorch, tensorflow and keras in terms which implements traditional machine learning algorithms ( Linear regression , knn, decision trees, ...
0 votes
1 answer
15 views

How to optimally visualise hyperparameter tuning?

I am working on a basic Neural network and want to show performance of model with respect to different parameters. I need help with suggestions for impactful visualizations. I can not make ...
-1 votes
0 answers
12 views

Counting the number of parameters in CNN

I could not find the answers of the following questions. Q.1. You have an input volume of 32 x 32 x 3. You want to process time-series data with a 1D CONV with zero padding, stride of 1, and 2 filters ...
0 votes
0 answers
15 views

Proof that averaging weights is equal to averaging gradients (FedSGD vs FedAvg)

The first paper of Federated Learning "Communication-Efficient Learning of Deep Networks from Decentralized Data" presents FedSGD and FedAvg. In Federated Learning the learning task is ...
0 votes
1 answer
23 views

Given an image how to find height of an object?

If I have an image of apple then how can I find the height of an apple using Deep learning? The photo of an apple is taken from the top view and I want to detect the height of that apple. How to do it?...
1 vote
0 answers
25 views

Extract categories from text (unsupervised)

I have a list of bank statements. They look pretty much like this: Received transaction. Reason[separator] invoice from [date][separator][number]. Counterparty[separator] [company name] Payment to ...
  • 155
1 vote
3 answers
82 views

How to learn steep functions using neural network?

I am trying to use a neural network to learn the below function. In total, I have 25 features and 19 outputs. The above image shows the distribution of two features with respect to one of the outputs....
  • 61
0 votes
0 answers
21 views

class label is less than 1 percent in classification problem

I am working on a ML problem where one class label is very less than even 1 percent. i.e 0.0002% I have tried undersampling, oversampling, SMOTE but the results are not satisfactory on the model. I ...
  • 101
0 votes
0 answers
9 views

How to build Stacked Autoencoder using Keras?

Stacked Autoencoder I have tried to create a stacked autoencoder using Keras but I couldn't do the last part of this autoencoder. Here I have created three autoencoders. It works fine individually ...
  • 1
0 votes
0 answers
30 views

CNN age classification --- low accuracy

I have a dataset of 34k (200x200) images and I want to build an 8 class age detector. I've tried a lot of different networks design, regularizations, dropout layers, grayscale images, data ...
0 votes
1 answer
19 views

Reusing a model, pretrained on 19 classes, for just one of those classes

I have a pretrained net for semantic segmentation, which has been trained on the cityscapes dataset and its 19 classes (Person, car, traffic sign, …). One of those is "Person". I am only ...
  • 1
0 votes
0 answers
29 views

Unable to load a .h5 keras model

I trained a multi-class Unet model using keras built-in metrics and loss (Accuracy and categorical_crossentropy). After the end of the training, I saved the model with the following command: ...
  • 111
2 votes
1 answer
43 views

Training Neural Network using TensorFlow on Large Video Dataset for Human Activity Recognition

I am working on a Human Activity Recognition on videos using deep learning and want to train my Neural Network (ConvLSTM & LRCN) on my custom dataset. The issue that I am facing is that my dataset ...
2 votes
0 answers
35 views

Should credit be given to AI model - low data scenario [closed]

In my office, we recently built an AI model for project success prediction using binary classification. Though the dataset size was small (977 records), my boss still wanted to go ahead with the POC ...
  • 2,337
0 votes
0 answers
12 views

What can I infer if large negative penalties are not increasing?

I am running a Deep RL algorithm. I defined a custom reward function. I run the algorithm for at least 500 epochs. For each epoch, I am printing the total reward received by the actor-network. It is ...
  • 147
2 votes
1 answer
42 views

How to remove noise from signals?

I have sensor which outputs signals (two signals bellow for example). I use 2000 signals as my data, which some of them are clear and some of them are bad signals. All clear signals have peaks, and ...
0 votes
0 answers
16 views

how to reduce the loss and improve the gradient flow - CNN

I am trying to improve this situation, in image classification[3 classes, softmax in the last layer], I constructed the neural network having 7[conv2d+Batchnormalization] layers + 1 linear layer, ...
0 votes
1 answer
22 views

How to increase FER2013 dataset validation_accuracy for only 3 classes i.e, happy,sad,neutral?

I am building a face emotion detection model using vgg16. Using FER2013 dataset for 7 classes i am getting= train_accuracy=97%, validation_accuracy=90%. but when i tried with 3 classes i.e, happy,sad,...
0 votes
0 answers
17 views

Captum vs GNNExplainer for explainability in Graph Neural Networks

I'm new to Graph Neural Networks and interested in exploring frameworks that allow the identification of nodes/edges that underlie prediction. I came across : (1) a model architecture (GNNExplainer) ...
  • 101
0 votes
1 answer
48 views

Classification on severe Class Imbalance high dimensional data

Dear DataScience Community, I am working on class imbalance tabular data with high-dimension inputs. The tabular data is derived from the satellite data pixels, and I have inflated the train data ...
1 vote
1 answer
24 views

Modelling Player Impact on English Premier League

Let's say you have a very wide feature vector, all 0s and 1s, all 500 players in Premier League as features, out of which only 22 participate in a match. Those that participate are marked with 1s, ...
2 votes
1 answer
23 views

Ignore or predict padding

I have a sequence to sequence classification model with two classes (similar to NER transformer) and because my data samples have different lengths I use padding. Is it better to use a custom loss ...
0 votes
0 answers
20 views

Are Soft and Bahdanau attentions different?

I have been working on a Image Captioning model. And read many articles accomplishing it. Some used both attentions interchangably while some did not. And the formulaes differed too. So, I would like ...
0 votes
0 answers
22 views

Triplet + Classification loss for Person-Reid

I am trying out a Person-Reid model on a custom dataset using Triplet Loss + Classification Loss (with label smoothing) on a custom dataset. Following are the configurations and the graphs: 1. Triplet ...
0 votes
1 answer
13 views

What actually is model size scaling and how do i globally apply to every model?

I read this article on the EfficientNet paper and have seen a lot of this kind of scaling. For example, there's Tiny-YOLO, YOLO (the base),.. Some model like SVTR, people did scale it to Tiny, Small, ...
0 votes
1 answer
19 views

Encoding Colour Output for a Sequential Neural Network

I’m fairly new to deep learning and AI and my first proper project consists of a model taking 13 or so inputs and one colour output. Currently, I’ve got 3 outputs (RGB) but I was wondering if I’d be ...
1 vote
0 answers
5 views

Issue with Relation Annotation (rel.manual) in Spacy's Prodigy tool

I am trying to build a relation extraction model via spacy's prodigy tool. NOTE: ner.manual, ner.correct, rel.manual are all recipes provided by prodigy. (ner.manual, ner.correct) The first step ...
0 votes
1 answer
22 views

Encoder Decoder model for parameter extraction from text input

I have an input as text from which I want to extract parameters as given in example below. Input: ...
0 votes
0 answers
13 views

Training and validation accuracy extremely low for Autonomous Lane Navigation via Deep Learning

Using a SmartCar running on RPI4 i collected all the images necessary for training. Training is done using CNN Nvidia's Model with Tensorflow and Python. Took about 900 Images for Up and 800sh for the ...
0 votes
0 answers
6 views

image classifcation model's depth and width

I wonder how deep and wide deep learning model should be. Where can I possess some information/rules how many layers and how wide they ought to be? I created basic image classification model with ...
  • 1
0 votes
0 answers
14 views

Why is my LSTM prediction is saturated and have bad prediction?

I am new to deep learning. Currently, I am trying to predict torque based on its past values using an LSTM model. There are two datasets (generated from a scaled test), one with wear and the second ...
1 vote
1 answer
24 views

Is backpropagation applied every layer the same?

For example, I have layers that are pretrained. But while predicted, the loss is very high. But not because of pre-trained layers. Because of not pretrained layers. Will every layer be affected by ...
  • 89

1
2 3 4 5
93