-
# Few ideas in here:
Stale gradients: Do async updates with a learning rate relative to the timestamp attached to a gradient. Also applicable for time series
Use rocks db for local synchronization …
-
I have the following program that fetches three JSON resources from an `https` server:
```
{-# LANGUAGE OverloadedStrings #-}
{-# LANGUAGE ScopedTypeVariables #-}
import qualified Data.Aeson …
-
what would it take to implement it?
like so:
`# ls /storage/videos`
```
video1.mp4
video2.mov
```
`# ffmpegfs /storage/videos /mnt/ffmpegfs`
`# find /mnt/ffmpegfs`
```
/mnt/ffmpegfs/…
-
Hi, I‘m trying to compress my Unet network which for regression. I just use 1-lambda*l2_loss for Acc, and till now trained 40 roll outs,the Acc reach only 0.22 but the Acc of original model is 0.96, i…
-
I am attempting to train model on AVA myself and faced very low quality of predictions. Digging further, I found predictions to be very strange and started to investigate pretrained models and asking …
hcl14 updated
5 years ago
-
Hi. I think Horovod is a data parallel distributed training framework. Since it is a sync framework, the gradient first averaged via ring-reduce and then broadcast to each device to update parameters …
-
Thanks for doing this, it's always good to have a quantized comparison in addition to still-image screenshot comparisons.
However, I have a few questions and suggestions:
1. Could you evaluate `ra…
-
Can we use [HashedNets](http://arxiv.org/abs/1504.04788) or [deep compression](http://arxiv.org/abs/1510.00149) to make it easier to distribute the parameters for our model to end users?
-
> AI - 人工智能;AR - 增强现实;CV - 机器视觉;DL - 深度学习;DM - 数据挖掘;DS - 数据科学;DV - 数据可视化;IOT - 物联网;ML - 机器学习;NLP - 自然语言处理
-
## Description
The `nav` section of `mkdocs.yml` contains an entry for `null`:
```yml
- not: ref/not.md
- null: ref/null.md
- or: ref/or.md
```
### Expected behavior
…
ghost updated
5 years ago