Train smarter AI models by learning to safely and securely use distributed private data with differential privacy, federated learning, and encrypted computation techniques. Apply for a Challenge Scholarship for the chance to participate in a vibrant student community and to earn one of 300 follow-up Nanodegree scholarships to continue your education with PyTorch and other deep learning tools.

Learn more about this challenge here

Challenge has been finished? Don't worry you can still join the course because its free. Join from here

I have already completed Introducing Differential Privacy and Evaluating the privacy of a function.

Today:

• I took the pledge.
• I am struggling with Deep Learning with Pytorch. Tried grasping video 12 and 13 of this lesson.
• Was confused in neural networks so took help from Andrew NG and completed the following videos from his famous Machine Learning course and gave his quiz and got 80%.
• Learned how logic operators can be represented as neural networks from Andrew NG.

By the end of Day 1, I have my nano-degree 36% viewed. Spent 3+ hours on Day 1.

• Learned the basic intuition of cost function and back propagation in neural networks

Even though Back propagation makes me wanna cry, I managed to get something out of it. Dived deeply into the derivation to find out what's really going on behind. Thanks to this article

Learned that how Chain rule is a powerful tool. Back propagation is just merely a tool to calculate the derivative of error with respect to weights for gradient descent. Correspondingly, gradient descent resulting in updating weights such that it minimizes the error.

OMG I did it with Pytorch! <3 Lalalala. Previously, few months ago I had learnt to train neural networks using Tensorflow but I hadn't try to go in-depth then. Now when I understand much of the concepts and its maths and how pytorch has made them easier to work with, I understand the power of neural networks.

What I did today:

• Watched an hour+ long lecture on back propagation, as suggested by udacity course mate, not to forget many other lectures, I have been trying to grasp about back propagation lately. https://lnkd.in/dvkJ2hm
• Revised video 12 and 13 of Lesson 2 i.e Intro to Pytorch, which were sounding alien before and got amazed how coming back to the same lecture after few days, meanwhile youtubing and googling about Back propagation, loss functions and gradient descent, strengthened my concepts and I easily learned the concepts of video 12 and 13 without pausing in between.
• Finally, attempted the following exercise.

• Built and trained a neural network for Fashion MNIST dataset.
• Watched 14,15 and 16 video of Lesson Deep Learning with Pytorch.

I want to understand how dropout works and a coursemate suggested me the 7th video. but for that, first, I need to learn the first 6 I have watched first three videos and 4th is in the midst. Tried to read this research paper on dropout first but left it in between xD

• Did revision of half completed deep learning with pytorch.
• learned about regularization from this source

Learned about dropout. At first it seemed to me like a funny technique, which nobody knows why works, but it does. But the "intuition" titled videos of Andrew NG always clarifies the remaining confusions, and loved the idea of drop out. How an algorithm shouldn't take any unit in layer for granted.

List of videos I watched today:

1. Why Regularization Reduces Overfitting (C2W1L05)
2. Understanding Dropout (C2W1L07)
3. Dropout Regularization (C2W1L06)

• Watched 17 and 18 lesson of Deep Learning with PyTorch Lesson.
• Coded practically, how over fitting leads to variance.

• Learned how dropout decreases variance
• Learned the difference between training a neural network for Mnist dataset between pytorch and tensorflow from this source

• Watched first 5 videos of convolutional neural networks Playlist and another video: Convolutional Networks

• Tried setting up Cuda environment in local machine by installing cuda toolkit of nvidia and download visual studio. torch.cuda.is_available() still showing False.

• Attended today's seminar of Bobby Wagner
• Learned the types of differential privacies and their difference in between (Lesson 5 first two videos)
• Again tried to run and troubkeshoot Cuda in my local machine. Not successful so far :(

• Finally got my Cuda working.
• Fell in love with my GTX 960 M <3 Realized that it trains my batch of 64 images 182 times faster than my CPU :D :D :D This example can also solve confusions of my peers, who wonder what CUDA is or why they should buy CUDA supported machines or how beneficial it is to buy CUDA supportted laptops.

• Watched last 8 videos of lesson Deep learning with Pytorch

• Watched videos 2 - 6
• Wrote my first ever kaggle kernel. This is about Randomized response in differential Privacy

I think the kernel is incomplete yet. I need to do de-skew the results maybe.

• Even though I had watched videos from 2 to 6 yesterday of local and global differential privacy lesson, I hadn't understood them fully. So today, I read people answers on slack about my confusions regarding these I posted yesterday and today on slack and practiced about the 60 70 % story in local and global differential privacy lesson.
• Updated the kaggle kernel with the (new code)[https://lnkd.in/dFisB_x]

1. Participated in Slack by asking questions: Question 1: https://lnkd.in/dTAyuyf Question 2: https://lnkd.in/dkj8m-p
2. Practiced local differential privacy code with noise. Proposed changes in the github python notebook because i feel like there is a mistake. https://lnkd.in/dXhKy9T
3. Attended #sg-pakistan virtual meetup
4. Attended #sg_pytorch-robotics virtual meetup

• Watched last 3 videos of lesson 5
• Participated in slack

• Watched first 5 videos of lesson 6

Learned about Reinforcement learning

• Watched this video: An introduction to Reinforcement Learning
• Read this article: Not-So-Deep Reinforcement Learning for dummies— Part 1

• Loved the concept of Federated learning. Too excited to do project on it. Watched first four video of this lesson.
• Setting up torch environment on Pi for #sg_pytorch-robotics group project. It is taking forever to install. Made swap partition. Installed python 3.6. Now am leaving it pytorch to install overnight.

• Continued on with installation of torch in pi for #sg_pytorch-robotics group project. Yesterday night I had put it on installation overnight but installation got messed up and my operating system broke and I had to reinstall it and do it all over again. I am leaving it on installation again today overnight. Hoping for the good. :( <3

• I was supposed to install pytorch in python 3.6 or above and I had mistakenly did in 3.5. so tadaaa again installation overnight in pi for #sg_pytorch-robotics :(
• Revised the first four videos of lesson federated learning.

• Successfully Installed torch and pysft in Raspberry pi and laptop for a group project. It was a challenging task to install in raspberry pi. The installation took 24 hours to complete, ignoring the fact that another 24 hours I had wasted by installing the above packages in the wrong python version.
• Practiced a bit federated learning basic coding
• Meetup with #sg_pytorch-robotics Sending thanks and encouragement to today's attendees: @Bhanujeet Choudhary@Elena Kutanov @Mohammad Diab @Dammy @Joyce Obi@Joyce @Marwa@erinSnPAI @Oudarjya Sen Sarma @Jess@Mateusz @qursaan @Suraiya Khan @Anju Mercian@Sourav @Ahmed MAGDY EISSA @Ayesha Manzur@Nancy Al Aswad @Droid @Laura A @susyjam @Harkirat@Vipul Aggarwal @Helena Barmer
• Participated in slack Q1: https://lnkd.in/dV4BmXW Q2: https://lnkd.in/dJMNeSY

• Watched this video about Federated Learning: Machine Learning on Decentralized Data (Google I/O'19)

Trying to work on a little task given at the end of the video 4 of lesson federated learning where we are required to make 2 virtual workers and play with it but this error keeps popping up. Reinstalled different versions of torch and syft to make thus error go away but nothing happens.

• Installed pytorch + pysyft in my another laptop + another pi (in process. Leaving it to install overnight). This time I just SSD instead of SD card with Pi and concluded that, before pi would take 12+ hours to build pytorch and with SSD it took only 3 hours. I would suggest everybody in #sg_pytorch-robotics to go for SSD instead of SD cards. I am all ready for project.

• Learned cloning SD card, instead of installing Everytime packages in a new SD card.
• Installed Jupyter notebook on Pi.
• Did coding on my Pi and pushed it to GitHub
• Watched this amazing video of How Neural networks look like in 3D

• Practiced basic remote execution
• Played with remote tensors by sending and receiving data from two virtual workers
• Learned doing arithmetic operations on remote tensors Practice Code

• Learned calculating gradient on remote tensors
• Learned building a simple linear model with pysyft
• Had a Zoom meetup with #sg_pytorch-robotics
• Sending thanks and encouragements to #sg_pytorch-robotics group members who joined today @Helena Barmer @Juan Carlos Kuri Pinto @Pooja Vinod @Jess @Bhadresh Savani @Oudarjya Sen Sarma@Elena Kutanov @Temitope Oladokun @Sourav @Joyce @Stanislav Ladyzhenskiy @Suparna S Nair@Nirupama Singh @Mateusz @Lisa Crossman @Ebinbin Ajagun @Mahitej @susyjam @Divya @Molly-ngampit @Dammy @Ayesha Manzur @Sayed Maheen Basheer @Mahak Bansal @Tanat Tonguthaisri
• Did a sneak peak in the AWS deep racer course Github: https://lnkd.in/dKUdAaA

• Learned about garbage collection in pysyft
• Toy federated Learning Practice Code

• Bought power supply for raspberry pi for group project #sg_pytorch-Robotics
• Read some articles and watched a video about federated learning:

1. Private AI — Federated Learning with PySyft and PyTorch
2. Federated Learning: The Future of Distributed Machine Learning
3. Making every phone smarter with Federated Learning

Watched

1. How Does Deep Learning Work? | Two Minute Papers #24
2. Federated Learning of a Recurrent Neural Network on Raspberry PIs
3. Encrypted and Decentralized AI - TensorFlow and Deep Learning

• Talked to Andrew Trask (yayyyy), regarding a problem in pysyft.
• Completed Federated Learning Lesson (except the project)
• Showing off 30 days of udacity badge
• Participated in Slack
• Watched the following research paper summary videos:

1. Google's DeepMind AI Just Taught Itself To Walk
2. 3D Style Transfer For Video is Now Possible!
3. AI Creates Facial Animation From Audio | Two Minute Papers #185
4. This AI Learned To See In The Dark

• Revised Federated Learning and did coding and pushed it to github
• Looking for a dataset, which could satisfy federated learning project + #sg_pytorch-robotics project
• Participated in slack
• Sending thanks and encouragemnets to #sg_pytorch-robotics mates: @Temitope Oladokun @qursaan @Shivam Raisharma @Jess @Sankalp Dayal @Nirupama Singh @Sushil Ghimire @Oudarjya Sen Sarma @Pooja Vinod @Juan Carlos Kuri Pinto @Alex Ahumada @Vigneshwari @Shashi Gharti @Helena Barmer @Nachiket @cibaca @Nirupama Singh @Dammy @Molly-ngampit @Elena Kutanov

• Trying to work on Federated Learning Project :(

• Completed Federated Learning Project with MNIST dataset, having two virtual workers
• Attended #sg_pytorch-robotics meetup
• Participated in slack
• Sending thanks and encouragements to: @Juan Carlos Kuri Pinto @Helena Barmer @Mika @Suparna S Nair @Nancy Al Aswad @Temitope Oladokun

• Worked on a toy dataset with a trusted aggregator
• Trying to understand Secretive additive Sharing

• Read some articles on secret additive sharing

• Uninstalled and Installed several times pytorch,syft,raspbian from 11 am to 11 pm, to get things right for #sg_pytorch-robotics project. Million thanks to Suparna for helping me out.

• Day 42
• Spent more or less 10 hours in configuring, installing things on Pi (Loving Raspbian Bluster) + training RNN for #sg_pytorch-robotics project. My laptop slept at 96% 🙂 🙂 🙂 I will have to go all over again ) P.S. Happy Independence Day to all Pakistanis

Done with the #sg_pytorch-robotics/showcase-challenge project

• After 3 failed attempts:

1. At 96% (laptop restarted)
2. At around 30 % (laptop restarted)
3. At 60 % (unstable internet) Because of unstable internet and my laptop sudden restarting for unknown reasons.
   Learned how to train a Recurrent Neural Network (RNN) in a federated way with the purpose of classifying a person's surname to its most likely language of origin.

Trained two Recurrent Neural Networks residing on two remote workers(Raspberry Pis) based on a dataset containing approximately 20.000 surnames from 18 languages of origin, and predict to which language a name belongs based on the name's spelling.

Published article on Medium about Federated Learning of a Recurrent Neural Network for text classification, with Raspberry Pis working as remote workers

Done with Secured Federated Leaning lesson

Revised Secured federated Learning

• Celebrated sir Andrew commiting to sharing my medium blog I wrote for #sg_pytorch-robotics group project at twitter @openminedorg.
• He also gave 50 claps on my article
• Mr. DanyEle, who is the author of the official blog at openmined also appreciated it.
• Completed the course
• Participated in slack P.S I feel like, I have mistakenly mentioned wrong day numbers on slack 60daysofudacity channel. I am sorry for that.

Scanned the following books vaguely to pick book to follow, since my Udacity course has been finished.

• Learned about ROS actions and services and practically used it in simulator
• Watched this video about Machine Learning: Living in the Age of AI | A WIRED Film.

Studied about how to use rotating proxies and user agents in web scrapping (for a personal digital assistant). I need the other web scrapping practitioners in my friend list to suggest me, how do they avoid to get blocked from websites doing web scrapping?

1. Do you use free rotating IP addresses and User agents or
2. do you use any free library which automatically handles the job of rotating IPs and user agents.
3. do you use any paid rotating IP addresses and user agents or
4. a paid library, which automatically handles rotating IPs and user agents. -- Learned about Unsupervised Learning from Andrew NG Machine learning course

• Learning about CNN from "Intro to deep learning with pytorch" course on Udacity.

Continued on with CNN Lesson in "Intro to Deep Learning with Pytorch Course" (Idk why they are revising MLP)

• Published my previously written article for review to "Secure and Private AI Writing Challenge"

• #sg_pytorch-robotics meetup. Thanks for joining: @Jess @Mika @Pranjal Chaubey @Elena Kutanov @cibaca @Mateusz @Stanislav Ladyzhenskiy @Sourav @Suparna S Nair @Helena Barmer. I encourage you all to create exciting new learning materials in the last days of this challenge.
• Learned following things in CNN

• Watched the video for suggested machine learnkng books
• Scanned the book "Machine Learning for Humans" it's good.
• Spent manyyy hoursss on updating the GitHub readme.md for maintaining the 60daysofudacity document. I was procrastinating it for so long. These books will help you learn machine learning youtube video

Trying to install Opencv in widows but why is it so difficult to do so. Was following this source I will try tomorrow with ubuntu. It will install in Ubuntu in no time i believe. I would encourage Helena Barmer and @Suparna S Nair to keep going :) problem solved: it was as easy as pip install opencv-contrib-python

Applied different vertical and horizontal filters on different pictures using opencv for "Intro to deep learning with pytorch course" on udacity I would like to motivate our Machine Learning PHD Scholar @Juan Carlos Kuri Pinto and skater girl @Mika to keep going :D

Deep learning cars were attracting me so I watched several videos and read articles about their implementation. Deep Learning for self-driving cars medium article

Ride in NVIDIA's Self-Driving Car

NVIDIA Jetbot Neural Network Based Collision Avoidance Demo

I would like to motivate @Frida @Jess to keep going as they have completed their 60 days.

The community has been awesome. I wanted to learn neural networks for so long, but without having nobody to solve my questions, I would always leave. Bidding the course mates farewell who have their last day today here

• Learned how to code CNNs in course "Intro to deep learning with Pytorch"
• Showed off 3rd position badge for Showcase project Challenge. Checkout the project here
• Connected with 90+ linkedin, same for Github and kaggel Udacity scholars via Linkedin, Github, Kaggle initiatives respectively on slack.
• Participated in last AMA session and reverse AMA session Note: I will fill in the 60daysofudacity completion form on 1 september to record my highest activity, and luckily i will be completing my 60 days on 1 September too. Documentation for 60dayofudacity I would like to encourage @Sayed Maheen Basheer and @sourav kumar to keep going even after this challenge :D

Did a bit revision of "Intro to deep learning with pytorch" course and learned further. I would like to encourage Suparna S Nair and Taimur Zahid to keep up their learning spirit after this challenge ends

Continued on with "Intro to deep learning with pytorch" course. Last day with this scholarship challenge 👻. I would encourage Aisha Javed and Taimour Abbas, my Pakistani mates, to keep up their learning spirits :)

I continued this streak during my final exams of 6th semester 😈. Learned a lot from the community. I didn't know until today, how truly environment affects you. Made soo many new friends. Dared to approach people like Andrew Liam Trask and Daniele Gadler personally in slack. Didn't know before, people like them are approachable. Online meetups,webinars, project and blogging competitions were fun. and the best thing is, it doesn't stop here. It revealed to me many more opportunities 🏆

Wishing myself luck for phase 2! I look forward to join Karachi Facebook Developer Group (which is inactive these days and will become active again soon) and keep an eye on Facebook AI residency applications opening :)

Got the badge