Questions tagged [mnist]
MNIST is a database of handwritten digits collected by Yann Lecun, a famous computer scientist, when he was working at AT&T-Bell Labs on the problem of automation of check readings for banks. The data set is a benchmark widely used in machine learning research. Due to its small size it is also widely used for educational purposes.s based on C++ but provides APIs for Python and Matlab.
96
questions
0
votes
1
answer
28
views
hacky backprop outperforms clean backprop - Why?
I implemented a basic NN for MNIST in Numpy and started with a hacky implementation of backprop (just randomly multiplying gradients together), but somehow that one works better than my cleaned up ...
0
votes
1
answer
25
views
My keras CNN model gives good predictions in 10/11 classes but missleads to the 11th class. What can I do to improve?
My project involves classifying printed numerical characters from real-life essays. My dataset includes 11 classes ('0' - '10'), with the label '10' representing the '/' symbol.
The issue is that ...
0
votes
0
answers
8
views
I wrote a classifier in C++ but it doesn't learn and returns the same loss each epoch
I wrote a classifier in C++ and tried training it on MNIST set but it doesn't learn. Because I am using log loss, it returns a loss of -ln(1/10) basically random chance. I tried tinkering with my ...
0
votes
1
answer
27
views
How to update first layer weights?
I’m trying to make a neural network without using any deep learning library that recognizes numbers in the mnist database. Its structure is: 784 input neurons (for the 784 pixels in the number images),...
0
votes
0
answers
21
views
Autoencoders failing to recreate MNIST numbers
I have been having trouble trying to get a working (non-variational) autoencoder to reproduce images from the MNIST dataset. The two biggest issues is an averaging of the samples to yield a single ...
0
votes
0
answers
16
views
Convolutional Network not Fully Learning MNIST (60% accuracy)
I have been working on a self-made Conv Net for some time. In addition to my network taking multiple epochs to increase the test-set accuracy, the accuracy flatlines at about 60% around 30 Epochs.
...
0
votes
0
answers
29
views
How to improve Accuracy on dermaMNIST dataset?
Unlike the regular MNIST which gets 97-99% with a fairly basic network, dermaMNIST gets training/validation stuck on 0.69. This tells me the model is underfitting. But, making it bigger seems to have ...
1
vote
0
answers
54
views
Negative MNIST - Trying to understand what's going on here
I'm new to machine learning and pytorch and any insight would be greatly appreciated. I've been playing around with the MNIST number recognition dataset and thought it would be interesting to train ...
1
vote
2
answers
35
views
Using a Genetic Algorithm in junction with a digit classifier CNN to create an "MNIST image generator"
I'm trying to use a genetic algorithm that optimizes a 28x28 matrix (its shape) to make it look like an image of the number 7 that could be found in the MNIST image dataset.
My attempt is to basically ...
0
votes
0
answers
38
views
GAN/DC-GAN isn't converging
I've been trying to train a vanilla GAN(for MNIST) for a few days, and nothing works. I've tried a lot of different layers, hyperparameters, and more, but every time the discriminator's loss decreases(...
0
votes
0
answers
11
views
Assessing performance of a generative model (images) using Jensen-Shannon Divergence
Background
I am currently working on comparing two high-dimensional image datasets (MNIST) of shape $\sim(50000, 256)$ to evaluate the performance of a generative model, and have attempted to use ...
2
votes
0
answers
77
views
Detecting upside down numbers in the MNIST dataset
I'm trying to predict numbers using the MNIST dataset. Since I don't know whether the numbers that I'm trying to predict are rotated (most of the time upside down, but can also be 90 or 270 degrees), ...
1
vote
0
answers
153
views
Is there a list of all the incorrectly labelled MNIST images along with their correct labels?
It seems that its well known that the MNIST handwritten digit dataset contains quite a few examples where the labels are clearly wrong and correspond to the wrong digit, some examples of people ...
0
votes
1
answer
28
views
I cannot run MNIST MWE (hello world for DL)
I have installed Anaconda and want tor run MWE for MNIST but I'me getting this error:
D:\STAZENE_last\Anaconda2\Lib\site-packages\torch\cuda\__init__.py:107:
...
0
votes
1
answer
42
views
What is the highest possible prediction accuracy when I flip some labels at random?
I want to predict MNIST labels in a binary setting using a simple MLP model (0 for digits 0-4 and 1 for 5-9). For the train and test data, I randomly flip 25% of the labels.
Is the maximum achievable ...