blocksparse
Efficient GPU kernels for block-sparse matrix multiplication and convolution (by openai)
AOgmaNeo
Arduino and Desktop compatible fast online/incremental machine learning system. (by ogmacorp)
blocksparse | AOgmaNeo | |
---|---|---|
1 | 7 | |
1,003 | 49 | |
0.6% | - | |
0.0 | 9.8 | |
11 months ago | 1 day ago | |
Cuda | C++ | |
MIT License | GNU General Public License v3.0 or later |
The number of mentions indicates the total number of mentions that we've tracked plus the number of user suggested alternatives.
Stars - the number of stars that a project has on GitHub. Growth - month over month growth in stars.
Activity is a relative number indicating how actively a project is being developed. Recent commits have higher weight than older ones.
For example, an activity of 9.0 indicates that a project is amongst the top 10% of the most actively developed projects that we are tracking.
Stars - the number of stars that a project has on GitHub. Growth - month over month growth in stars.
Activity is a relative number indicating how actively a project is being developed. Recent commits have higher weight than older ones.
For example, an activity of 9.0 indicates that a project is amongst the top 10% of the most actively developed projects that we are tracking.
blocksparse
Posts with mentions or reviews of blocksparse.
We have used some of these posts to build our list of alternatives
and similar projects. The last one was on 2021-01-27.
AOgmaNeo
Posts with mentions or reviews of AOgmaNeo.
We have used some of these posts to build our list of alternatives
and similar projects. The last one was on 2021-01-27.
-
[Discussion] Is there any alternative of deep learning ?
My own work focuses on an alternative to deep learning, called Sparse Predictive Hierarchies (SPH). It is implemented in a library called AOgmaNeo (Python bindings also exist). It does not use backpropagation, runs fully online/incremental/continually (non-i.i.d.). Its main advantages are the online learning but also that it runs super fast. Recently, I was able to play Atari Pong (with learning enabled!) on a Teensy 4.1 microcontroller, and still get 60hz.
-
Is deep rl possible on microcontrollers?
Yes, but the standard techniques are too computationally heavy to learn directly on an arduino. You might try something alternative such as OgmaNeo https://www.youtube.com/watch?v=Zl6Rfb3OQoY https://ogma.ai/ Or otherwise you need to offload the learning computation onto a powerful CPU/GPU.
-
RL model getting good and then bad
Our software's user guide (describes our software which does fully online learning without forgetting)
-
[P] Real2Sim Interactive Demo with AOgmaNeo
If you want to know more about how AOgmaNeo works, here is a guide.
-
Can you make this in TensorFlow or PyTorch? (AGI sketch)
Sounds like OgmaNeo2. They use integers for your point number 5. If an integer variable can hold numbers between 0 and 19, which means 20 possible values, then its real value is always exactly 5% of the possibilities. Use several of these in a layer to handle combinatorial explosion in the input data. The advantage is that integers as a substitute for one-hot encoded booleans take away a whole dimension, which means that they are damn fast. So fast that https://github.com/ogmacorp/AOgmaNeo even runs on an Arduino! I don't think this is possible with TensorFlow or PyTorch.
-
[D] Has anyone here gotten any advantage in inference/training time, or memory footprint by using sparse models instead of Dense? Theoretically, sparsity sounds efficient, but I haven't personally able to get a slight boost even with 90x sparse Resnets.
Not really deep learning, but our biologically-inspired online learning system called AOgmaNeo makes heavy use of sparsity. It can run at over 200fps with 800,000 synapses on a Teensy 4.1 (Arduino-compatible microcontroller). It can also recall 256x256 video 1 minute long on a desktop CPU (trains at ~60 fps).
-
[P] Quadruped Reinforcement Learning with AOgmaNeo
Repository Python Bindings Ogma Website
What are some alternatives?
When comparing blocksparse and AOgmaNeo you can also consider the following projects:
Lorcan-Mini - Smaller version of my previous quadruped robot, Lorcan!