-
When we initially implemented the crypto PAL layer for Android, we implemented it entirely in JNI as the implementation in Android is only accessible in Java and we didn't have a mechanism to write Ja…
-
## Describe the bug
In paper "Noisy Networks for exploration" they say (section 3.1) "A noisy network agent samples a new set of parameters after every step of optimisation. "
But in current imp…
-
We need to create some combination of the current quality layers to make a simpler mask for users to apply. Possibilities include
- `(conncomp == 0) & (temporal_coherence > 0.5)`
- `(conncomp == 0…
-
part of (after deduplication)
-
how are `top-level-pins ` made in KLayout?
there are `MetalN.pin`, `TopMetalN.pin` and `GatePoly.pin` layers. but there are also the `*.text` layers.
is a top_lvl_pin for the LVS-script a combinati…
-
Error occurred when executing KSamplerAdvanced:
free_upper_bound + pytorch_used_bytes[device]
-
Currently Haliax requires that all names in a single named array be unique. In general I think this is a good constraint. However, for Linear layers it's frequently a nuisance, since one often project…
-
What method do you think would be best for noising the LSTM?
End of Noisy Network paper seems to suggest that randomization technique from "Bayesian recurrent neural networks" https://arxiv.org/abs/1…
-
In https://github.com/kbembedded/msemu/pull/35#discussion_r331761742 it was suggested to expand logging capabilities to have different layers.
e.g.
log_debug() would be very very verbose
log_tra…
-
Hi there,
I am currently using the brevitas weight_bit_width quantisation for quantisation aware training on a convolutional SNN.
Since this method outperforms my other QAT methods, I am wondering…