-
## URL(s) with the issue:
[negative sampling](https://www.tensorflow.org/tutorials/text/word2vec#negative_sampling_for_one_skip-gram)
[generate training data](https://www.tensorflow.org/tutorials/te…
-
Here are two [very particular] examples of contributing a tool to a major oss project, `scipy`. They are both examples of tools that have a scientific computing flavour — tools that have some deeper m…
-
_From @shorowit on May 4, 2017 20:40_
PAT 2.0 includes an algorithm setting for Number of Samples when using Diagonal sampling, but this value is not actually used. Rather, the number of datapoints i…
-
Potential readings to focus understanding on probability, mathematical statistics, algorithms, etc more deeply. Talk of applications will be secondary.
- [Statistical modeling: the two cultures](ht…
-
`-max()` might be zero probability, but it's still better than `-infinity()`. Accept samples if that happens. See
```
const bool trial_sample_has_zero_probability
= ((trial_lo…
-
Sort of adversarial autoencoder. Train a normal autoencoder (with constraints on the latent space to prevent drifting) simultaneously with a normal GAN (or perhaps an LSGAN?).
-
all algorithms, e.g. local thresholder should take the calibration of images into account.
currently, the local threshold on xyz would typically use a much too big z-range compared to xy.
-
## Background
The current implementation has a probability limit of `1 / (2^32-1)` for the integer constructor and a limit of `1 / (2^64)` for the f64 constructor. Assuming a couple nanoseconds per…
-
-
Model arithmetic is a great codebase for combining signals from different LLMs and discriminators. Given the trend of [doing search at inference time](https://arxiv.org/pdf/2408.03314v1) (such as MCTS…