Arduino Neural Networks

Neural Networks are another type of Artificial Intelligence used in computing. They are used in computer games, expert systems and at many more places.

A in a limited way you are able to use them on Arduino too - eg. to steer an Arduino robot! In my presentation I will explain more about this topic.

This slides were presented on my presentation "Arduino, roboti a neurální sítě" at Czech Arduino Day 2015 on BarCamp Plzeň (more info at https://plzenskybarcamp.cz/2015/arduino-day) #ArduinoD15 #Arduino #barCampCZ

Transcript

1. Tomáš Jukin @Inza Arduino and Neural Nets
3. 16 MHz
4. 2 KB RAM 16 MHz
5. :-(
6. :’-(
8. ? ? ?
9. 7 + 8 + 4 = 19!
10. 7 + 8 + 4 = 19
11. 7 + 8 + 4 = 19 Input Hidden Output
12. 7 + 8 + 4 = 19 7 Segment Display to Binary
13. Tomáš Jukin @Inza www.juicymo.cz @JuicymoCZ
14. Tomáš Jukin @Inza academy.juicymo.cz @JuicymoAcademy
16. Dendrite Nucleus Axon
17. Input Activation Function Output
18. Dendrite 2 Nucleus Axon Dendrite 1
19. Input 2 Activation Function Output Input 1
20. Input 2 Activation Function Output Input 1 0,2 0,8 0,43 0,18 0,37
22. Input Hidden Output Feed Forward ANN
23. Input Hidden Output
24. Input Hidden Output Ultrasonic Sensor Gyroscope Wheel Encoder Left Motor Right Motor
25. Input Hidden Output Ultrasonic Sensor Gyroscope Wheel Encoder Left Motor Right Motor Flow
26. Input Hidden Output Ultrasonic Sensor Gyroscope Wheel Encoder Left Motor Right Motor 20º = 20/180 = 0,111…
27. Input Hidden Output Ultrasonic Sensor Gyroscope Wheel Encoder Left Motor Right Motor 20º = 20/180 = 0,111…
28. Input Hidden Output Ultrasonic Sensor Gyroscope Wheel Encoder Left Motor Right Motor 20º = 20/180 = 0,111…
29. Input Hidden Output Ultrasonic Sensor Gyroscope Wheel Encoder Left Motor Right Motor 20º = 20/180 = 0,111… 0,345 = 34% power
30. Input Hidden Output Ultrasonic Sensor Gyroscope Wheel Encoder Left Motor Right Motor 20º = 20/180 = 0,111… 0,345 = 34% power 0,4 = 40% power
31. Input Hidden Output Ultrasonic Sensor Gyroscope Wheel Encoder Left Motor Right Motor 20º = 20/180 = 0,111… 0,345 = 34% power 0,4 = 40% power 0,4 - 0,345 = 0,055
32. Input Hidden Output Ultrasonic Sensor Gyroscope Wheel Encoder Left Motor Right Motor 20º = 20/180 = 0,111… 0,345 = 34% power 0,4 = 40% power 0,4 - 0,345 = 0,055
33. Input Hidden Output Ultrasonic Sensor Gyroscope Wheel Encoder Left Motor Right Motor 20º = 20/180 = 0,111… 0,345 = 34% power 0,4 = 40% power 0,4 - 0,345 = 0,055 W11 = 0,14 W12 = 0,65 W14 = 0,42 W13 = 0,26 W21 = 0,21 W22 = 0,37 W24 = 0,19 W23 = 0,46
34. Ultrasonic Sensor Gyroscope Wheel Encoder Left Motor Right Motor 20º = 20/180 = 0,111… 0,345 = 34% power 0,4 = 40% power 0,4 - 0,345 = 0,055 W11 = 0,14 W12 = 0,65 W14 = 0,42 W13 = 0,26 W21 = 0,21 W22 = 0,37 W24 = 0,19 W23 = 0,46 Backpropagation
35. Ultrasonic Sensor Gyroscope Wheel Encoder Left Motor Right Motor 20º = 20/180 = 0,111… 0,345 = 34% power 0,4 = 40% power 0,4 - 0,345 = 0,055 W11 = 0,14 W12 = 0,65 W14 = 0,42 W13 = 0,26 W21 = 0,21 W22 = 0,37 W24 = 0,19 W23 = 0,46 Backpropagation
36. Ultrasonic Sensor Gyroscope Wheel Encoder Left Motor Right Motor 20º = 20/180 = 0,111… 0,345 = 34% power 0,4 = 40% power 0,4 - 0,345 = 0,055 W11 = 0,14 W12 = 0,65 W14 = 0,42 W13 = 0,26 W21 = 0,21 W22 = 0,37 W24 = 0,19 W23 = 0,46 Backpropagation
37. Ultrasonic Sensor Gyroscope Wheel Encoder Left Motor Right Motor 20º = 20/180 = 0,111… 0,345 = 34% power 0,4 = 40% power 0,4 - 0,345 = 0,055 W11 = 0,14 W12 = 0,65 W14 = 0,42 W13 = 0,26 W21 = 0,21 W22 = 0,37 W24 = 0,19 W23 = 0,46 Backpropagation
38. When to use ANNs? It can learn and detect patterns
39. When to use ANNs? But beware local optimum!
40. When to use ANNs? To much training is bad!
41. When to use ANNs? To much training is bad! To few training is bad too!
42. And what about hardware?
43. How ANN stores data? FF ANN needs just a two dimensional array
44. How ANN stores data? FF ANN needs just a Feed-Forward Artificial Neural Net
45. Nice! So I can run it everywhere then, can I?
47. ANN on UNO? NOPE :-(
48. ANN on UNO? 2 KB RAM => 19 neurons effectively
49. ANN on UNO? Average ANN has >100 neurons
50. ANN on UNO? Interesting ANN has >2000 neurons
51. ANN on UNO? => No go on UNO …
53. ANN on Raspi? Yop, that is a start…
54. ANN on Raspi? Yop, that is a start… and I2C or UART will connect it with UNO…
55. ANN on Raspi? Yop, that is a start… and I2C or UART will connect it with UNO… but BEWARE the VOLTAGE!
56. Case Study
57. #Probee Robot
58. #Probee Robot 3x UltraSonic Sensor (Forward / Backward / Turret) 4x DC Motor 1x Motor Controller (Left / Right) 1x Arduino UNO (= brain) 1x Bluetooth Module 1x 2x16 I2C LCD (= status display)
59. #Probee Robot UNO ANN Hours to learn movement…
60. #Probee Robot UNO ANN Hours to learn movement… 15 minutes battery
61. #Probee Robot UNO ANN Hours to learn movement… 15 minutes battery Needs better HW preparation … =>
62. Interested? Build or Implement start today! You can start right now… …with any language… …or on Arduino!
63. Inspiration http://goo.gl/sKTfsd Blog post about ANN on UNO http://goo.gl/cuIUiH ANN controlling robot in video
64. Questions?
65. Tomáš Jukin @Inza www.juicymo.cz @JuicymoCZ
66. Photo Credits All photos used are CC from Flickr !