Someone told me that neural nets work great with Dense data like images but for sparse data they are not very good. For sparse data sets you go with traditional algorithms like logistic regression, SVD etc. Is that a known thing or perhaps something that could have been a single off instance of the person’s experience?
If you work with sparse data you can embed it to make it dense. For example, let’s says your input feature is a category with 1000 classes. The input data with be very sparse (one-hot vector of size 1000). But the embedded vector could be a dense vector of size 10.