ECE 490/590
Neural networks
Vikas Dhiman
Barrows Hall Rm 105,
[email protected]
## Vikas Dhiman (he/him/his) #### Career + BTech in Electrical Engineering + Worked as an IT Engineer for 4 years + MS in Computer Science + PhD in Robotics #### Personal + Hobbies: hiking, skiing, biking, video games, books on politics/sociology.  
Terminology
## ImageNet moment <img src="/ECE490-Neural-Networks/assets/0000-00-03-intro/imagenet-challenge.jpg" style="width:700px"/> 
## The Turing awardees 
## Similar courses at UMaine + ECE 491/591: Deep Learning (Dr Yifeng Zhu) - 25-30% overlap but we will dive deeper+ ECE + COS 498/598: Machine Learning (Dr Salimeh Y Sekeh) - 25-30% overlap but we will be focused on Neural Networks
## Similar courses elsewhere 1. Machine Learning Specialization (2022) [Website](https://www.coursera.org/specializations/machine-learning-introduction) | Instructor: Andrew Ng 2. [DEEP LEARNING NYU Fall 2022](https://atcold.github.io/NYU-DLFL22) Instructor : Alfredo Canziani, Yann LeCun 3. [Deep Learning in Computer Vision with Prof. Kosta Derpanis (York University)](https://www.eecs.yorku.ca/~kosta/Courses/EECS6322/) 4. [Stanford CS231n](https://cs231n.github.io/); Instructors: Fei-Fei Li
## Syllabus [vikasdhiman.info/ECE490-Neural-Networks](https://vikasdhiman.info/ECE490-Neural-Networks)
## Demos 1. [Hugging face Object detection](https://huggingface.co/spaces/CVPR/Object-Detection-With-DETR-and-YOLOS/discussions) 2. [Image captioning](https://huggingface.co/spaces/flax-community/image-captioning) 3. [Image generation](https://huggingface.co/spaces/stabilityai/stable-diffusion) 4. [Hugging face](https://huggingface.co/spaces) 5. [ChatGPT](https://chat.openai.com/auth/login)
## Limitations What Neural Networks can not do? ## Ethics What Neural Networks should not do?
## Ethical concerns because of high-accuracy * Face recognition * Deepfakes[^1] [^1]: https://www.cs.princeton.edu/~arvindn/talks/MIT-STS-AI-snakeoil.pdf
## Fundamentally dubious * Predicting criminal recidivism * Predicting job performance * Predictive policing * Predicting terrorist risk * Predicting at-risk kids[^1] [^1]: https://www.cs.princeton.edu/~arvindn/talks/MIT-STS-AI-snakeoil.pdf
## Case of lie detectors * "Overall, the cumulative research evidence suggests that when used in criminal investigations, the polygraph test detects deception better than chance, but with error rates that could be considered significant."<br/>  * Keeps coming back: "Tracy Harpster, had ... a miracle method to determine when 911 callers are actually guilty of the crimes they are reporting."<br/> 
## Thank you * Prereq homework coming soon
