top of page
Blogs I Latest News
Use tab to navigate through the menu items.
Data challenge application
Select a data challenge
Note : The DIAGNODENT challenge is exclusively available to EU citizens due to GDPR regulations and the confidentiality of the dataset.
Your teammates emails
Country of residence
Why do you want to respond to this challenge (intentions, goals) ?
Samples of AI projects you already deploy
Assume a multi-label classification containing different types of images (photos, x-rays, etc.). An image could contain one or more labels. The dataset is highly unbalanced. Describe the solution you are trying to implement and why you think it is the best approach?
in the big data context, one might need to use reduction techniques to exploit large scale structures (high resolution images). in this context explain briefly the main idea behind the use of reduction techniques
What is the total number of parameters, including bias, in a single 1x1 convolution filter when the input image is of size 64x64x16?
Your model is underperforming during training. The last hidden layer outputs very very small values.What do you do?
Add a residual connection
Add more images to your dataset
Increase the shape of your input layer
Reduce the dropout
Which of the following techniques can be used to address the issue of exploding gradients in deep learning models?
Utilizing the SGD optimization method
Oversampling the minority classes in the dataset
Increasing the batch size during training
Imposing gradient clipping during the backpropagation process.
Feature maps at the beginning of a CNN, as opposed to those towards the end of the network:
have semantically stronger features
can be freezed during learning transfer
cover spatially finer areas of the image
Training phase shows no improvement in the model after first epochs:
increase learning parameter
increase the size of the model
verify loss function
Subscribe to Our Newsletter
Thank you! We will get back to you soon !
bottom of page