1 of 33

Responsible TinyML

Brian Plancher

Barnard College, Columbia University

brianplancher.com

1

CC BY-NC-SA 4.0

2 of 33

How can TinyML support Responsible AI?

2

CC BY-NC-SA 4.0

3 of 33

How can TinyML support Responsible AI?

3

Accessibility / Education

Sustainability / Conservation

Privacy / Security

CC BY-NC-SA 4.0

4 of 33

Accessibility / Education

4

CC BY-NC-SA 4.0

5 of 33

Promoting Accessibility / Education

5

TinyMLedu.org

CC BY-NC-SA 4.0

6 of 33

Promoting Accessibility / Education

6

TinyMLedu.org

CC BY-NC-SA 4.0

7 of 33

Global Embedded ML Education Opportunities:

7

Low Resource Requirements

1

Low Power

Low Cost

Low Connectivity

Interdisciplinary Focus

and Applied Learning

2

CC BY-NC-SA 4.0

8 of 33

Global Embedded ML Education Opportunities:

8

Low Resource Requirements

1

Interdisciplinary Focus

and Applied Learning

2

Low Power

Low Cost

Low Connectivity

CC BY-NC-SA 4.0

9 of 33

Global Embedded ML Education Opportunities:

9

250 Billion

MCUs today

Software and Hardware Fragmentation

1

Challenges

CC BY-NC-SA 4.0

10 of 33

Global Embedded ML Education Opportunities:

10

Language

and Local

Relevance

Software and Hardware Fragmentation

1

Affordability Barriers and Localization Roadblocks

2

Challenges

CC BY-NC-SA 4.0

11 of 33

Global Embedded ML Education Opportunities:

11

Software and Hardware Fragmentation

1

Affordability Barriers and Localization Roadblocks

2

Educator Readiness and Research Incentives

3

Challenges

CC BY-NC-SA 4.0

12 of 33

12

CC BY-NC-SA 4.0

13 of 33

Towards a Modular Curriculum

13

CC BY-NC-SA 4.0

14 of 33

Towards a Modular Curriculum

14

Please feel free to remix our materials and please consider sharing back your materials for the community!

TinyMLedu.org

CC BY-NC-SA 4.0

15 of 33

Calls to Action

15

Assessing Our

Educational Programs

1

Maintaining Open-Source Software and Courseware

2

Embedded ML Model

and Data Zoo

3

Improving Accessibility

of Hardware

4

Growing a Research Community

5

Increased Outreach

and Diversity Efforts

6

CC BY-NC-SA 4.0

16 of 33

Calls to Action

16

Assessing Our

Educational Programs

1

Maintaining Open-Source Software and Courseware

2

Embedded ML Model

and Data Zoo

3

Improving Accessibility

of Hardware

4

Growing a Research Community

5

Increased Outreach

and Diversity Efforts

6

CC BY-NC-SA 4.0

17 of 33

Calls to Action

17

As has been noted in related works, this kind of methodology has many flaws and does not take into account much of the nuance in gender, including issues of bias, misperception, and nonbinary identities [7], [8]. However, we hope that this initial study will help add to the robotics community’s understanding of the current state of gender diversity and, at a minimum, provide directionally correct data to help with future diversity, equity, and inclusion efforts.

ieeexplore.ieee.org/document/10474552

Underrepresentation of Women in Robotics Research

CC BY-NC-SA 4.0

18 of 33

Sustainability / Conservation

18

CC BY-NC-SA 4.0

19 of 33

Promoting Sustainability / Conservation

19

TinyMLedu.org

CC BY-NC-SA 4.0

20 of 33

20

TinyML can support the SDGs but comes with costs. What is the net impact?

CC BY-NC-SA 4.0

21 of 33

Building Representative Systems

21

High Cost

Medium Cost

Low Cost

Image Classification

Keyword Spotting

Cost Level

Application

Size

Large

Compact

CC BY-NC-SA 4.0

22 of 33

22

harvard-edge.github.io/TinyML-Footprint/

CC BY-NC-SA 4.0

23 of 33

TinyML Systems in Context

23

5x to 38x Savings over a 3-year lifespan!

CC BY-NC-SA 4.0

24 of 33

What if we scale to 250bn devices?

24

https://www.statista.com/statistics/1183457/iot-connected-devices-worldwide/

There are around 250bn MCUs deployed today and around 15bn IoT devices

CC BY-NC-SA 4.0

25 of 33

What if we scale to 250bn devices?

25

If our devices have longer lifespans this data starts to look even better!

CC BY-NC-SA 4.0

26 of 33

Privacy / Security

26

CC BY-NC-SA 4.0

27 of 33

IoT 2.0: Intelligence on Things

27

Google Assistant

nest

IoT 1.0:

Internet

of Things

TinyML will soon be everywhere!

CC BY-NC-SA 4.0

28 of 33

What is a Machine Learning Sensor?

28

Sensor 2.0

Privacy by Design

mlsensors.org

CC BY-NC-SA 4.0

29 of 33

We suggest transparency as a core value

29

arxiv.org/abs/2306.08848

mlsensors.org

CC BY-NC-SA 4.0

30 of 33

Materiality and Risk in the Age of Pervasive AI Sensors

Brian Plancher

Barnard College, Columbia University

brianplancher.com

30

arxiv.org/abs/2402.11183

CC BY-NC-SA 4.0

31 of 33

Evolution of Sensors…

31

CC BY-NC-SA 4.0

32 of 33

… and their impact on Responsible AI

32

Valid and Reliable

Safe

Secure and Resilient

Accountable and Transparent

Explainable and Interpretable

Privacy Enhanced

Fair with Harmful Bias Managed

CC BY-NC-SA 4.0

33 of 33

How can TinyML support Responsible AI?

Brian Plancher

Barnard College, Columbia University

brianplancher.com

33

Accessibility / Education

Sustainability / Conservation

Privacy / Security

CC BY-NC-SA 4.0