We have been testing the ability to release daily builds for sometime and now are ready to make it available to all the users. Daily build is the mechanism which we have chosen, to make all the latest features and fixes available to the users without waiting for an actual release.
Daily builds can be found at : /nfs/ncl/daily-builds on your INRC Cloud VM. The naming convention for daily builds is : next_release_version-daily-year_month_date. We will be updating the Release Notes with the new features, updates and bug fixes so that you know the changes made since the last stable release.
As these builds are being pushed out on a daily cadence, they might not be available every day due to build/regression failures, might contain partially implemented features and/or might have unchecked bugs. If a certain daily build does not work for you, try to pick a previous one or wait for the next daily build to fix any high priority blockers.
As always, if you find an issue, please raise a ticket by email : nrc_support@intel-research.net
We will continue putting stable releases at /nfs/ncl/releases for major milestones.
Mike Davies and Garrick Orchard presented at the CVPR 2019 Event-based Vision and Smart Camera Workshop in Long Beach CA. The presentations and slides are now publicly available. The workshop website contains full presentations and slides from all speakers.
We are pleased to announce the release of NxSDK 0.8.7. The installation files are located at /nfs/ncl/releases. Installation steps stay the same as of 0.8.5. Please refer to getting started guides within Startup Materials[also available in nxsdk-apps-0.8.7.tar.gz/docs]
The release is available for all participants who have signed the INRC participation agreement and have access to the INRC Cloud.
New Features/Improvements
NxNet
epl: EPL module now works for multi-pattern learning and has an updated Jupyter notebook tutorial which shows how the EPL network can learn and recall not only odors but also images. EPL is a neuromorphic one-shot learning algorithm which can learn and recall patterns using a spatio-temporal attractor network inspired by the neural circuitry of the external plexiform layer (EPL) of the mammalian olfactory bulb.
Python Tutorials | Feature Covered | Description | Category |
|---|---|---|---|
tutorial_25_sequential_host_snips | Host Snips | This tutorial demonstrates how to setup sequential host snips, provide implementation and scheduling and connect them to embedded snips via channels | Snips |
NxCore
N2Board and N2Chip now expose various methods for allocating chips and cores on demand (not only during N2Board setup) at manually or automatically determined mesh locations. Manually determine mesh locations can be specified by (x, y, p) address or logical core index.
General
Board supports creation of host snips in sequential and concurrent mode. See documentation for further detail on host snips api.
Improved live DVS support. Users can now write their own snips to receive live DVS spikes and inject them into their model.
API Changes/Deprecations
board.createProcess API is being deprecated. Use board.createSnip instead.
board.n2Chips[k].n2Cores now returns a dict mapping the logical core index to the n2Core object instead of a list of n2Core objects. This is to support the ability to allocate cores at arbitrary mesh locations. A list of allocated n2Core objects can be accessed through board.n2Chips[k].n2CoresAsList.
net.createDVSSpikeGenProcess API is being deprecated. Use the DVS module instead. See the DVS module tutorial for an example.