Scribd
Upload
475 views14 pages

TinyML in The Developing World PDF

The document discusses how TinyML, or machine learning on low-power embedded devices, may be well-suited for applications in the developing world. It provides examples of current TinyML projects like detecting plant diseases and illegal deforestation. Embedded devices require less infrastructure than smartphones and can operate without networks or power grids. Their low capital costs make education and development accessible. The author believes TinyML could help address acute problems in the Global South regarding water, agriculture, health and ecology. However, they want to avoid a "White Savior complex" and empower local experts by providing educational materials on new tools, and learning what needs experts in the field have.

Uploaded by

dame diongue
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
475 views14 pages

TinyML in The Developing World PDF

The document discusses how TinyML, or machine learning on low-power embedded devices, may be well-suited for applications in the developing world. It provides examples of current TinyML projects like detecting plant diseases and illegal deforestation. Embedded devices require less infrastructure than smartphones and can operate without networks or power grids. Their low capital costs make education and development accessible. The author believes TinyML could help address acute problems in the Global South regarding water, agriculture, health and ecology. However, they want to avoid a "White Savior complex" and empower local experts by providing educational materials on new tools, and learning what needs experts in the field have.

Uploaded by

dame diongue
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 14

TinyML in the Developing World

[email protected]
Who am I?

Technical Lead of TensorFlow Lite Micro at Google

- Open source machine learning framework


- https://www.tensorflow.org/lite/microcontrollers
- Fits machine learning models in less than 20KB
- Used for embedded devices
- Contact me at [email protected]
What am I here to talk about?

- TinyML is the idea that machine learning on cheap, low-power embedded


devices opens up new applications
- I believe it has a lot of characteristics that make it a good fit for uses in the
developing world
- I’m here to share what I know, and then learn and listen from domain
experts to better understand if there are good uses
Practical Examples
Plant Village
Collaboration between PennState
and Google

https://news.psu.edu/story/5132
36/2018/04/02/research/penn-s
tate-developed-plant-disease-app
-recognized-google
Rainforest
Connection
https://blog.google/technology/ai/figh
t-against-illegal-deforestation-tensorflo
w/
There are a lot more examples

https://www.wildlabs.net/ is a fantastic resource to learn more

The examples I showed are all using smart phones


How is TinyML different?

Smartphones Embedded

Large solar panels Tiny panels or AA batteries

Bulky, hard to install Size of a coin, peel and stick

Expensive, $100+ Cheap, < $1

Sensitive to heat and moisture Hardy

Easy to program Hard to program


What sort of things can you do?

- Detect audio: Chainsaws, voice interfaces, bird calls


- Detect movement: Gesture recognition, magic wands, packages
- Detect images: Count people, cars, animals, inventory
- Other sensor data: Monitor machinery, power usage
How do you learn to program these devices?

- You can buy an embedded development kit for $15


- There are cheap or free learning resources:
- https://www.edx.org/professional-certificate/harvardx-tiny-machine-learning
- https://tinymlbook.com
Why do I think this might fit developing
countries?

When there’s no legacy infrastructure, a lot of innovation happens, as seen with


smart phones.

Embedded devices can work with no network or power grid.

The capital costs of education and building devices are very low.

A lot of problems that embedded ML can address are acute in the Global South,
including water, agriculture, health, and ecology.
Am I Right?
What are the next steps?

- I want to avoid White Savior complex. I’m very ignorant of what people
actually need.
- My goal is to empower the domain experts in the developing world who are
facing these problems.
- I hope to provide educational materials to help them understand what new
tools have emerged over the last few years.
- I’m hoping to hear from those in the field about what they need.
Thanks!
Contact me:

[email protected]
@petewarden

You might also like