-
Implementing the Neural Tangent Kernel adaptive loss method proposed in the "When and Why PINNs Fail to Train: A Neural Tangent Kernel Perspective" [paper](https://arxiv.org/pdf/2007.14527.pdf) by Sif…
-
Hello! I was wondering if it's possible for one to write their own `kernel_fn` in neural tangents to do regular kernel regression (for example, a Gaussian kernel). Naively just writing a `new_kernel_f…
-
Can we use neural tangent kernel [1] to get the benefits of new neural net architectures while using the budgeted kernel machines? See also the paper on the path kernel [2].
The kernel needs to com…
-
# Quanta | science, reflection, projects
undefined
[https://xyli1905.github.io/2019/08/neural-tangent-kernel-and-nn-dynamics/](https://xyli1905.github.io/2019/08/neural-tangent-kernel-and-nn-dynamic…
-
Hi developers, I've met a problem when using neural-tangents as follows:
```
KERNEL_FN = nt.utils.batch.batch(KERNEL_FN, batch_size=kernel_batch_size)
AttributeError: module 'neural_tangents' has…
-
# Abstract
The analysis of over-parameterised Artificial Neural Networks (ANNs) reveal that the optimisation (training) process only slightly changes the parameters of the model. This allows one to a…
-
## 一言でいうと
幅が無限とした場合のDNNを最急降下法で学習させる処理は、線形変換と同等であるとした論文(=学習結果は一次のテイラー展開で置き換えられる)。損失関数が2次の場合は出力がGaussianで維持されるため学習の過程はGaussian Processと見做すことができ、この点がBayesianNetと異なるとしている。
### 論文リンク
https://arxiv…
-
Great job! I wish that I had seen this earlier.
I would like to recommend
1. [When and why PINNs fail to train: A neural tangent kernel perspective](https://arxiv.org/abs/2007.14527)
2. [On the e…
-
Hey!
I was curious about why you choose this as the batch size:
https://github.com/google/neural-tangents/blob/9f2ebc88905c46d60b7c4a9da25636924acc9d45/neural_tangents/utils/batch.py#L611
I'm n…
-
I have used this library to calculate jacobians for the [Neural tangent kernel](https://en.wikipedia.org/wiki/Neural_tangent_kernel) study.
It's fast and reliable and I'm just thinking to **extend th…