Jeremie Desgagne-Bouchard

```julia (CatDogPanda) pkg> st Project CatDogPanda v0.1.0 Status `C:\github\CatDogPanda\Project.toml` [336ed68f] CSV v0.10.9 [052768ef] CUDA v3.12.1 [88a5189c] DataAugmentation v0.2.11 [587475ba] Flux v0.13.11 [916415d5] Images v0.25.2 [2913bbd2] StatsBase v0.33.21 [5e47fb64] TestImages v1.7.1...

I think you're having the right diagnosis. I just tried: ```julia path = imgs[1] _img = testimage(path) _img = apply(tfm_train, Image(_img)) size(_img.data) img = channelview(float32.(_img.data)) julia> typeof(img) Base.ReinterpretArray{Float32, 3, RGB{Float32},...

Interesting PR @ablaom, the changes seem overall to move in a promising direction! Quick consideration regarding `predict`: this PR uses the prediction type (ex: `LiteralTarget`) as a second argument. My...

I had only initially experimented with Zygote, but would acutally opt for an Enzyme-based approach if I were to move forward with it. I haven't pushed much on this as...

That's an interesting case, if I'm understanding correctly, the problematic is not just about defining a new loss function as `f(pred, y)` (a function of only a model prediction and...

I also noted the efficiency issue raised regarding currently available Julia implementation of distance correlation: https://github.com/pfarndt/EnergyStatistics.jl/issues/2 What I understood from a quick look at `dcov` calculation is that it can...

Regarding the interface, the suppport of kwargs in `fit` such as `feature_names`, `weight_name` etc. could be expanded to support something like `decorrelated_feature_names`. The kind of direct support I'd bee contemplating...

My understanding is that DisCo didn't have to bother about higher order derivatives as it was trained as a regular neural network, and as such only needed first-order grad along...