Adam Hillier

I'm not experienced enough with C++ types and casting behaviour to really comment on what the right implementation of this would be, but I agree that it would be very...

> I'm confused as to why the end2end test works with `ReLu`. Ah yes, I should have mentioned this in the PR. Doing this transformation prevents ReLU fusing, because it...

> That's really cool, I completely forgot about this 🎉 > > In larq we already have the unipolar [`SteHeaviside`](https://docs.larq.dev/larq/api/quantizers/#steheaviside) quantizer (although I think the backward pass should actually be...

I have this working locally, and will make a PR to add support.

@rameshKrSah we haven't made any specific efforts towards implementing a binary dense layer. I think the most obvious and easiest way we could support binary dense layers would actually not...

Just to note, until we fix this in LCE this issue can be worked around by explitly adding TensorFlow fake quantise ops before and after the problematic layer that doesn't...

@lgeiger do you have any idea why a `Conv2D` combined with a `Conv2D` works though, wouldn't that run into the same issue of shared bias tensors?

I agree that matching directly against the pattern in LCE would be good. If it doesn't cause any issues, it might be nice to still keep the `experimental_implements` so that...

Hi @ideepankarsharma2003, at the moment you're right that the Keras [RNN](https://www.tensorflow.org/api_docs/python/tf/keras/layers/RNN), [LSTM](https://www.tensorflow.org/api_docs/python/tf/keras/layers/LSTM), and [GRU](https://www.tensorflow.org/api_docs/python/tf/keras/layers/GRU) layers have no binary equivalent in Larq, so I'll note this as a feature request for...

@bferrarini does your code work if you replace `DoReFa` with some other quantiser, such as `SteSign`?