Running CNN on ATmega328P

• larq

  2 692 7.5 Python

  An Open-Source Library for Training Binarized Neural Networks

  

You quantize the model parameters i.e., don't just send the model in which uses floating point math instead change it to fixed point. This has 2 advantages 1) a pure size reduction and 2) most low power MCU's don't have float point multipliers but do have single cycle fixed point multipliers. This is a classic DSP trick used for a long time. The real research aspects come-in as you start dropping below 8-bit; even coming down to single-bit in some cases(see Larq)

• InfluxDB

  www.influxdata.com featured

  Power Real-Time Data Analytics at Scale. Get real-time insights from all types of time series data with InfluxDB. Ingest, query, and analyze billions of data points in real-time with unbounded cardinality.

  

Suggest a related project