![CIIC 5995-100 / ICOM 5995-100
Human Perspective in Artificial Intelligence
(HPAI)
Professor José Meléndez, PhD
“We have accompanied the reader into the labyrinth of pattern recognition, pointed
out some of the salient landmarks, and offered battle on [their] behalf to the Minotaur
of semantic difficulties. We hope that we have given [them] enough strength to
encourage [them] to return.” (Adapted) - George Nagy (1937- )
Minotaur Skullcleaver, MTG
“State of the Art in Pattern Recognition”, Proc. IEEE, Vol. 56, No. 5, May 1968](https://image.slidesharecdn.com/hpai-class10-memoryintrospection-031120-200311232301/85/Hpai-class-10-memory-amp-introspection-031120-1-320.jpg)
Hpai class 10 - memory & introspection - 031120
- 1. CIIC 5995-100 / ICOM 5995-100 Human Perspective in Artificial Intelligence (HPAI) Professor José Meléndez, PhD “We have accompanied the reader into the labyrinth of pattern recognition, pointed out some of the salient landmarks, and offered battle on [their] behalf to the Minotaur of semantic difficulties. We hope that we have given [them] enough strength to encourage [them] to return.” (Adapted) - George Nagy (1937- ) Minotaur Skullcleaver, MTG “State of the Art in Pattern Recognition”, Proc. IEEE, Vol. 56, No. 5, May 1968
- 2. Next Up • Homework #4: • Due as scheduled on Wednesday March 18 by 11:59PM • Artificial vs Human Memory • Brief Introduction to Memory and the Brain • Artificial Memory • Digital Memory • Analog Memory • Human Memory • Mine and Yours – Round Desk Discussion • Functional Descriptions • Inside the Brain - Neurons, Glia and More
- 3. In what year was this written?
- 4. A Survey of Analog Memory Devices (on Moodle) • George Nagy, Cornell University, 1962. • https://www.ecse.rpi.edu/~nagy/PDF_chrono/1963_Nagy_Anal og_Memory_IEEE_EC1963.pdf
- 5. In what year was this written?
- 6. Neural Networks – Then and Now (on Moodle) • Letters, IEEE Transactions on Neural Networks, V.2, N.2, pp.316- 318, 1991. • Special-Purpose Parallel Analog Neural Networks • https://www.ecse.rpi.edu/~nagy/PDF_chrono/1991_Nagy_NN1 991-ThenAndNow.pdf
- 7. Modern Time
- 8. Required “Reading” • A Survey of Analog Memory Devices (on Moodle) • George Nagy, Cornell University, 1962. • https://www.ecse.rpi.edu/~nagy/PDF_chrono/1963_Nagy_Analog_Memor y_IEEE_EC1963.pdf • Neural Networks – Then and Now (on Moodle) • Letters, IEEE Transactions on Neural Networks, V.2, N.2, pp.316-318, 1991. • Special-Purpose Parallel Analog Neural Networks • https://www.ecse.rpi.edu/~nagy/PDF_chrono/1991_Nagy_NN1991- ThenAndNow.pdf • How Emotions are Made: The Secret Life of the Brain • Chapter 4: The Origin of Feeling
- 10. Allen Institute for Brain Science • As fascinating and inspirational as it is discouraging • 80B+ neurons and 80B+ glia – All Dead • 23 Chromosomes from a handful of human brains • 25,000 genes and other DNA – selectively sampled • Static and partial picture of a few humans. Conclusion: The brain is sufficiently complex to make us human!
- 11. Properties of Human Memory • Despite progress scientists don’t know: • how human memory really works. • how memories are encoded. • how memories are stored using clusters of neurons. • the role of non-neural brain cells in memory. • the partitioning or co-existence of processing and data.
- 12. Next Up • Artificial vs Human Memory • Artificial Memory • Digital Memory • Analog Memory • Human Memory • Mine and Yours – Round Desk Discussion • Functional Descriptions • Inside the Brain - Neurons, Glia and More
- 13. Digital Computer Memory • Storing: • Data or Instructions are “Encoded” into “1”s and “0”s for storage. • Cells or “Areas” are written to store “1”s and “0”s. • Retrieving • Cells or “Areas” are accessed to read the stored “1”s and “0”s. • Data or Instructions are “Decoded” for “Use”. • Duration: Short or Long Term • “Use” Includes: • Display (Analog) • Speaker Output (Analog) • Transmission (Digital – previously Analog) • Computation (Digital) • Restorage (Digital)
- 14. Digital Computer Memory – Compression Does the image file have to store all of the image pixel data?
- 15. Digital Computer Memory Advanced Compression Do both compressed files need to be saved separately?
- 16. Typical Computing Memory Types • Memories for Instructions and Data • Non-Volatile – Long Term Storage • Volatile – Short Term Memory • “Immediate” at CPU • ROM • Cache Source Unknown
- 17. CPU Cache – “Immediate” https://searchstorage.techtarget.com/definition/cache-memory
- 18. CPU Cache – “Immediate” • L1 • I for Instructions • D for Data • L2 • Mixed Instructions and Data • Shared by Multiple CPU Cores • L3 • Interface to DRAM https://cdn-images-1.medium.com/max/1200/1*nT3RAGnOAWmKmvOBnizNtw.png
- 19. Next Up • Artificial vs Human Memory • Artificial Memory • Analog Memory • Human Memory • Mine and Yours – Round Desk Discussion • Functional Descriptions • Inside the Brain - Neurons, Glia and More
- 20. Analog Electronics • The world is analog. • Analog is characterized by continuously variable signals • Electrical signals include: • Voltage • Current • Charge • Frequency • Phase • Significantly more “variable” than the digital world.
- 21. The Heart https://www.thoughtco.com/thmb/P6E8tnBs_Mp_JSKDlaRbbGNNvQA=/768x0/filters:no_ups cale():max_bytes(150000):strip_icc()/heart_inner_section-577d5c673df78cb62c939314.jpg • Right Atrium receives de-oxygenated blood from body xfers to Right Ventricle • Left Atrium receives oxygenated blood from lungs xfers to Left Ventricle • Right Ventricle pumps de-oxygenated blood to lungs • Left Ventricle pumps oxygenated blood to body
- 22. Electrocardiogram (ECG) https://www.researchgate.net/profile/Selvakumar_Gopalasamy/publication/228668 313/figure/fig1/AS:302084206415872@1449033895418/Normal-ECG-waveform.png • Measures electrical activity of heart through it’s contractions • Electrodes are attached in the chest area in a variety of positions relative to the heart • ECG signals have characteristic shape Example of Normal ECG CycleO is the origin or datum point preceding the cycle P is the atrial systole contraction pulse Q is a downward deflection immediately preceding the ventricular contraction R is the peak of the ventricular contraction S is the downward deflection immediately after the ventricular contraction T is the recovery of the ventricles U is the successor of the T wave but it is small and not always observed
- 23. Electrocardiogram – Normal or Abnormal? http://hqmeded-ecg.blogspot.com/2018/10/a-normal-ecg-on-busy-night.html Normal or Abnormal ECG?
- 24. Analog Functional Memories (ca. 1967) https://arxiv.org/pdf/1803.00485 • Physician Electrocardiogram Classification (ECG Only) – Normal (95%) v Abnormal (53%) • Using Trainable Hardware Nonlinear Matched Filters: • Electrocardiogram (ECG) Classification – Normal (97%) v Abnormal (90%) • Specht, “Generalization of polynomial discriminant functions for pattern recognition,” IEEE Pattern Recognition Workshop, Puerto Rico, October 1966. Example of an Analog Nonlinear Filter
- 25. ECG Frequency Spectrum https://i.stack.imgur.com/A7QJT.jpg Normal or Abnormal ECG?
- 26. Electrocardiogram – Normal or Abnormal? http://hqmeded-ecg.blogspot.com/2018/10/a-normal-ecg-on-busy-night.html Normal or Abnormal ECG? Analog EEG Signals Analog Matched Filter • High? – Normal • Low? – Abnormal Output
- 27. Analog Computer https://upload.wikimedia.org/wikipedia/commons/4/42/Bifnordennomenclature.jpg /a-normal-ecg-on-busy-night.html • 1945 Norden Bombsight • Opto-Mechanical Analog Computer • Designed to drop bombs accurately • Calculated bomb impact location based upon flying parameters