Skip to content

Releases: leondavi/NErlNet

Nerlnet version 1.5.2 - Neuropeptide

03 Aug 22:28
fb4fb1e
Compare
Choose a tag to compare

Nerlnet version 1.5.2 - Neuropeptide

  • Fix critical bug in Federated Worker.

Nerlnet version 1.5.1 - Neuropeptide

08 Jul 16:58
b280e56
Compare
Choose a tag to compare

Nerlnet version 1.5.1 - Neuropeptide

  • Fix bugs in Federated Worker.
  • Nerlplanner 1.0.3
  • Add distributed token to batch gathered info

Nerlnet version 1.5.0 - Neuropeptide

22 May 23:35
72104e7
Compare
Choose a tag to compare

Nerlnet version 1.5.0 - Neuropeptide

Release Notes

  • Introducing new worker to worker communication (W2WCom) module
  • FSM controllers are based on the new W2WCom
  • Federated Learning Avg. implementation based on W2WCom
  • Add Flatten layer to Nerlplanner
  • Fix issues
  • Introducing streams: Source to worker and Worker to worker.
  • Add Stream handlers (Start/End Stream messages)

Nerlnet version 1.4.2 - Endorphins

28 Mar 00:24
a2cfe5f
Compare
Choose a tag to compare

Nerlnet version 1.4.2 - Endorphins

  • Optimization of communication between main server and api server when sending results.
  • Optimization of NerlNIF call to train, remove redundant nerltensor conversion.

Nerlnet - Endorphins

24 Mar 13:43
3d4006a
Compare
Choose a tag to compare

Nerlnet version 1.4.1 - Endorphins

Fix critical issues that were found to support distributed experiments.
More information in the PR: #294

Nerlnet - Endorphins

13 Mar 22:49
5b9b498
Compare
Choose a tag to compare

Nerlnet version 1.4.0 - Endorphins

This version introduces new infra design to support various libraries of edge ML by setting a clear and easy API for translation of NN models parameters in NerlNIF.

  • Improve APIServer communication with MainServer
  • Introduce ApiServerDB, improved structured data base to collect ML results and Nerlnet cluster statistics.
  • Introduce SourcePiece for flexible management of allocated batches from CSV with multiple sources.
  • New experiment flow design (new exp json style)
  • Introducing experiment phases for flexible management of experiments
  • Add support for CNN layers
  • OpenNN version with CNN supported
  • Add support of AE and AEC
  • Improve communication statistics of all entities
  • Improve Stats class of ApiServer
  • Improve experiment flow API
  • Add Bounding Layer support to Nerlplanner and NerlworkerOpennn
  • Add Models NIF tests (Perceptron, CNN, AE)

Nerlnet - Vasopressin

14 Jan 21:45
60303fb
Compare
Choose a tag to compare

Nerlnet version 1.3.0 - Vasopressin.
Introduces NerlPlanner, an advanced UI to generated architecture, connections map and experiment flow JSONs. It allows researchers to create experiment on scale.

  • Fix bugs and improve stability.
  • NerlnetPlanner UI, first version 1.0.0
  • Introduce DC files and deprecation of ARCH files
  • Improved Json parsing (Erlang implementation)
  • Improved statistics views in API-Server
  • API Server severe issues were fixed.
  • Fix source sending frequency - Actual frequency is added
  • Improved logs.
  • Add epochs support to Source and Worker.
  • Add probability layers support
  • Add scaling/unscaling support to all layers
  • Add Pooling support to all layers
  • OpenNN v6.0.4 (Cpp14)
  • Introduce NerlWorker/NerlLayer generic classes of infra
  • Training strategy instance is saved between source-batches.
  • Routers are based on lookup table (Thanks @galhilu)
  • Communication of entities is based on Routers LUT solely

Nerlnet - Oxytocin

14 Jul 00:50
e46ed04
Compare
Choose a tag to compare
  • Fix high frequencies segmentation fault
  • Add explanation when a port is captured
  • Stability issues solved in NIF

Nerlnet - Oxytocin

15 Jun 21:48
4f45f8f
Compare
Choose a tag to compare
  • Fix bugs of high frequencies experiments
  • Fix Raspberry Pi issue of linking with -latomic
  • More improvements

Nerlnet Oxytocin - Efficient Communication and federated learning workers

12 Jun 18:31
179fb2e
Compare
Choose a tag to compare

Nerlnet - version 1.2.0 Oxytocin

  • NerlTensors are introduced in this version (Binary and Erlang)
  • NerlTensors can be encoded as double or float (binaries instead of Erlang lists).
  • Improve the efficiency of the communication network.
  • Add confusion matrix of workers.
  • Add generic worker and workerNN - Easier worker implementation in Erlang.
  • Improve NIF efficiency by using DMA copies from binary instead of list to tensor assignment cell by cell.
  • Improve logs.
  • Improved Json parser.
  • Tests and CI support.
  • Introducing the GenericWorker FSM - Intelligence is implemented on worker FSMs.
  • Implement workerFederated client and server.