- uTensor – Embedded Devices And Machine Learning Models
- BYOM – Bring Your Own Model
The conference took place in the QEII Conference Centre located at the City of Westminster opposite Westminster Abbey and nearby Big Ben (Elisabeth Tower), London Eye, Downing Street and Buckingham Palace.
uTensor – Embedded Devices And Machine Learning Models
uTensor converts a TensorFlow model into C++ code that can be executed on an embedded device. This allows machine learning models to run on an edge computer. The advantage is privacy, availability and speed. The data remains on the device and does not pass into the Internet, this is faster (depending on the model) and independent of any Internet connection. Within a CX context, these devices could, for example, determine whether a product has been taken off the shelf and put back. Recognition can be done via sensors or a camera, so possible images of the customers are not captured and all data will stay in the store.
The MNIST Machine Learning model recognizes handwritten numbers. The example shows a three that I drew and the result.
One question from the audience was if uTensor runs on a micro:bit or Calliope to teach children machine learning. Unfortunately uTensor needs the Mbed OS 5 from ARM. The micro:bit and Calliope only run Mbed OS 2, so uTensor cannot be used, but it is an interesting idea.
BYOM – Bring Your Own Model
In my book (Machine Learning mit SAP Leonardo) I looked into the Bring You Own Model (BYOM) topic. It is about how a trained model can be brought into the productive environment. I used SAP Leonardo Machine Learning Foundation and SAP Cloud Platform. As an example I used again a XOR model and a Fiori application with the Boston Housing data.
Some MCubed London talk
Other interesting talks at the MCubed were:
- Keynotes by Sebastian Riedel (FAIR) on Natural Language Processing Without Supervision. A highlight was the Muppet Revolution with ELMO and BERT.
- The second keynote Agile Applied AI: A New Discipline by Lorien Pratt (Quantellia Inc) had Decision Intelligence (DI) as its topic, which showed up in the 2019 Gartner Hype Cycle for Artificial Intelligence.
- Mobile AI and keeping humans-in-the-loop (Rupert Thomas) was about using AI to detect recyclable bottles. (His talk was based on my last years talk Machine Learning Models on Mobile Devices)
- Comparing Different Approaches To Text Analysis (Christian Winkler): ELMO and BERT
- PyTorch – From Research to Production? (Fabian Bormann): Machine Learning with PyTorch and C++.
It was a nice conference with nice talks, chats and the typical London rain. You can find my slides here: