How has your journey been so far, learners?


(James Requa) #164

Hi All,

I have been meaning to introduce myself properly here on the forums so sorry this is a little late :slight_smile:

A little bit about me…I’m an entrepreneur with a business degree and non-technical background. How did I get into DL? Well the story is that my business was really struggling and I didn’t have many options left (I have 2 kids so a lot of pressure :slight_smile: ), so I figured maybe if I pick up some more technical skills such as coding then I could come up with some solutions. Long story short is I ended up in the Deep Learning and AI nanodegrees at Udacity. Our fellow fastai students @Ekami and @apil.tamang were actually my classmates!

Since I don’t have any formal CS degree and at the time I was just getting started coding so it was a little daunting and overwhelming at first, but I found through a lot of repetition things started to slowly make sense. Also, it was while I was enrolled with Udacity that I discovered fast.ai part 1 MOOC videos and fell in love with the course and the teaching style. It really helped me to understand the underlying concepts of DL much more clearly. It was also fastai that motivated me to participate in my first Kaggle competition “Planet: Understanding the Amazon from Space” along with @Ekami as teammates haha…Now I am officially addicted - currently participating in 10 active competitions atm!

I also wanted to take this opportunity to wholeheartedly thank @jeremy and @rachel for making this course (my heart goes out to Rachel - hope she is on a speedy path to recovery!). Not only is the education top-notch but also I have found that fastai seems to attract some really amazing people from all over the world and the sense of community here is above any that I have experienced in other courses. I feel so fortunate to be here amongst all of you!

-James


(Apil Tamang) #165

@jamesrequa
Great introduction.

I appreciate your enterpreneurship (i can never spell this word right), and definitely admire how far you might’ve come. You’re definitely very active at the forums and such.

I’m a backend Java developer, and Python is (at times) very annoying. I hate it that I cannot always tell what kind of objects are being passed to a method because of the lack of static typing support for Python. I still loose my way through the ‘*’ operator, the zip(…), stack( ), lambdas, iter, next, and list operators, to name a few. Although now with a bit of getting used to, I feel better at it and definitely see the utility of those shorthand type operators.

I’m not sure what I hoped to get out of this program, but would like to be better at PyTorch, or so I thought. Although I’ve been trying to interface better with the fastai library, I hope to write some more native PyTorch code going forward. For right now, two evenings later and I’m still struggling to come up with the final piece for the planet kaggle competition to generate the submission file. Now if I could only understand what the opt_th(…) method does, and how to leverage that…

Good to share and know you all. Keep posting. I wish I could spend more time on this matter, and build and train SeNet, Capsule Nets etc. a full time job keeps me pretty much on the edge most days.

@jeremy @rachel and everyone who contributed to the fastai lib. and videos, really appreciate the effort and opportunity.

  • Apil

(Soumitra Kumar) #166

Firstly, thanks @jeremy and @rachel (wishing you speedy recovery!) for creating this course!

I am working from last ~20 years in variety of fields from transistor level simulations to big data application. Recently, I am working in Big Data Application Performance Management domain. I really like the top-down approach to teaching otherwise had to spend 5-10 years learning lots of theory before trying on DL! I am a visual learner, so like when things are explained in easy to understand visual way vs lots of equations and numbers.

I would like to use the techniques learned to do timeseries anomaly, and prediction. @jeremy please send me some pointers.


(nok) #167

Hi I am from Hong Kong. I have been watching videos for Part 1 and Part 2 v1. I start learning ML and DL last year from MOOC and gained some experience from Kaggle. I am planning to take it seriously and study v2 line by line.


(Tom Brown) #168

I’m just joining today now that the version 2 of part 1 has been unofficially released.

I’m a business analyst working in agile ways with cultural non-profit professionals to light their “dark data” in order to serve their patrons in better ways.

I’m interested in recommender systems when it comes to working with my cultural non-profit colleagues. On a personal basis as a digital photographer / image-maker, I’m particularly interested in the style transfer section of part 2.


(Cedric Chee) #169

Hi everyone,

Happy New Year!

I am from Singapore. My background is in software engineering. I quit my job early of Dec last year to focus on my study. I am currently transitioning to be a deep learning engineer at a startup working on medical diagnostic. Prior to that, I was working full-time in a local consulting company providing web and mobile app development services to our clients around the world. I am an ‘expert’ generalist and self learner/life-long learner.

I first got to know deep learning when I visited Stanford CS231n website where there’s a demo model training live in the web browser. Out of curiosity, I dig deeper and found it was created by Andrej Karpathy who was at that time the co-lecturer for CS231n. So, I started learning neural network in JavaScript through that demo. Strangely, I don’t understand a thing about neural network at that point of time. So, I keep looking and discovered Andrew Ng’s Stanford Machine Learning course and I completed it about 2 years ago. This give me enough grounding to study CS231n course where we learn to develop deep learning model for Convolutional Neural Network (CNN) from scratch by hand using just Python and numpy.

When Part 1 v1 of the course just released to the public, I took the chance to study and completed it. I didn’t manage to complete Part 2 v1 due to my full-time job obligations. With the “soft-launch” of Part 1 v2 and in conjunction to Nurture.ai’s AI Saturdays global initiatives kicking off this week, I am planning to re-take Part 1 v2 seriously together with some local study groups here. I think it’s worth doing. On top of that, I think we can learn PyTorch and the new fastai library. What not to love. :smile:

:two_hearts: Thank you to Jeremy and Rachel for the awesome course. :heart:

-Cedric


(emsart) #170

Hello, everyone and happy New year. Well I’m from France, with a background in econometrics and applied mathematics in the decision theory field. I work in the consulting industry for about…two decades now and came along this course from class-central.com I’m currently finishing the Deeplearning.Ai specialization and felt quite interested in learning how to deploy and execute models on the cloud, as well as crafting new implementations-models in other fields than strictly computer vision (even though image classifying is not per se a “vision” problem), mainly decision making and forecasting.

Like Cedric I’d like to thank Rachel and Jeremy for the great work and awesome course here.


(Ahmad Arib) #171

Hi everyone, my name is Arib from Indonesia, not found any Indonesian before, just one Malaysian (hi neighbor), please come forward fellow Indonesian if you guys read this.
I start Deep learning 1 year ago, in the big data class when I get to pair up with 1 Ph.D. guy from Pakistan (he’s not continuing this pursue sadly), he is the one who told me about deep learning and Tensorflow. We end up presenting our final project using Tensorflow. I get hooked up with deep learning, taking Andrew Ng class, taking formal machine learning class in school (I’m taking master in computer science at Shanghai, China), following another professor weekly lab meeting that work in deep learning, since my lab is about embedded & pervasive computing, following Stanford Conv Network online class, get stuck in implementation, found fast.ai. After a moment with fast.ai bought GeForce 750 Ti with pocket money (I’m a poor student), it so sucks, now working on my proposal to get my sponsor fund me buying 1080 Ti :smiley:
I implement simple computer vision in agriculture field in Indonesia to detect paddy pest and disease, get flight ticket back home from biggest telco company in Indonesia to join their social competition, that really simple idea that just runs on simple 750 Ti considered as quite a novel idea for my country, oh my, I get tons of idea ready to execute in more powerful machine, wish me luck with my sponsor fund. Thanks, @jeremy and @rachel without you guys I’m still hanging around with no clear direction how to implement the idea, haha.


(Ed Pureza) #172

Hello All,

I’m really excited to (re)start this course. I had just finished week 1 of v1 and took a little break over the holidays. I was surprised to see that there is a v2 and was glad to have taken a little break.

I majored in CS, but aside from 1 year of coding custom middleware components for banking web-enablement early on in my career, I had not had a technical role. My experience has mainly been in supply chain management (SCM) as a functional consultant and solution architect.

I recently finished the Coursera Machine Learning Specialization from the University of Washington. It was supposed to have a deep learning component which was unfortunately dropped from the specialization (course not yet developed). I really liked the set of courses with its case study based approach and was happy to find fast.ai and its approach to teaching to start my education in deep learning.

I’m still exploring applications of deep learning in my domain. So far, my list is composed of user activities that require decisions and manual intervention using current process (e.g., promotional activity classification, finding commonalities in items so that they can be managed together, what to do when there is not enough information such as when forecasting new items, identifying and managing exceptions like when a stock-out is projected or when demand changes drastically, etc.). I’m hoping going through these lessons will also help me see new ways of thinking about current SCM processes - perhaps a paradigm shift in how forecasts and replenishment plans are generated and managed or how supply chains will be orchestrated. I’ve seen other posts here that are related and I’m sure deep learning will be a big part of the next generation supply chain (if it isn’t already).

I liked reading about the different backgrounds of the learners here - so much important work to be done and discoveries to be made. Many thanks to @jeremy and @rachel and others for making this course available and making deep learning accessible to all.

Looking forward to learning!

-Ed


#173

Hi everyone,

I’m a software developer from Italy.

I’m grateful to @jeremy and @rachel for this course, and I’m also grateful to the community members for their support.

I had my first exposure to ML about two years ago, when working on a project that required a recommendation system (we used Apache PredictionIO with a customized recommendation engine template, based on a collaborative filtering algorithm).

Deep Learning is uncharted terrain for me, and I felt intimidated by the heavy mathematical background which is considered to be a prerequisite for even the simplest tasks, but the top-down approach on which the course is based is really a great idea and makes learning feasible and interesting.

Even though I can’t commit myself fully to the course due to work/family duties, I hope I’ll be able to learn as much as possible and give back too.


(Aseem Bansal) #174

Hi all

I always wanted to be a magician. Sounds naive but is true. As I grew up I was looking for real magic. To me software engineering sounded like magic. Creating something and having it work was magic for me. That got me into software. After I understood programming Machine learning sounded like magic.

I started learning with ML in mid-2014. That was just after I got my first job. I was trying out Andrew Ng’s course. Got bored. Stopped that. Tried again in mid-2015. Same thing happened. At the end of 2015 the CTO of the project that I was working for (a person whose experience is same as my whole age) said that we want to start with ML. He shared some coursera courses and said he would love people to jump in and do it. I thought that was the best thing ever as now I had the opportunity to align my work with this. So I tried and completed some of the courses. They were in R. They were better but still not so great. Then in beginning of 2016 I started looking and found Udacity. The Data Science Nanodegree. It had a monthly subscription and if I started doing that I knew I would have nothing left. But I decided this is important and took the step. 8 months later where I dedicated most of my spare time I completed it. It was good. The ML initiative in the project was looking bleak. But I learned.

In 2017 I found fast.ai randomly via a reddit thread. I don’t use reddit that much but it was just random that I found it. I think I was really lucky that day because it was an eye opener. I was like finally I found the magic that I was looking for. After Jeremey and Rachael’s explanation of bottom-up and top-down approach I finally understood why I was bored with the coursera courses and why I was not get anywhere. I did not complete all courses but till middle of Video 4 of Part 1 v1. But that was still great learning. Gave me a lot of confidence about things people consider magic. ML initiative in the company finally went to production. There were problems but the CEO had explicitly mentioned that exploring AI was really important to him. Others in the industry are doing it and reporting good returns so it is important. Currently working in production.

Middle of 2017 I got a chance to become Google certified Data engineer. I never would have cleared the ML/DL questions if I had not went through this course even half way. That’s when I understood I finally found the right thing. Later I got sick for a few months so wasn’t really able to focus on anything. I started with the computational linear algebra but that is still in the early parts.

I participated in a hackathon recently. The theme was hack for india. I tried my hand but wasn’t able to get anything good for solving local issues because of lack of data. But then I remembered that Jeremy mentioning somewhere that you don’t need a lot of data to get good results. I was thinking ok maybe I should complete the course. Maybe I need to discuss data science machine learning more. Because the current government of India is organizing many hacks. There is a vision for Digital India. Maybe learning about what is possible and then going out educating others and discussing with more people is what is required for me to be able to get to solutions. Coincidentally the company that I am working for also wants to get more people into this field and as I am the one who has been working on this they asked whether I would like to help with that and I said yes.

Today I saw http://www.fast.ai/2018/01/02/diversity-2018/ and decided that’s great because international fellowship will open and I will have a deadline. I know that when I have deadlines I do great. So made a resolution to fulfill the requirements of international fellowship and get both of the courses completed by end of April.

I believe in God and I believe that he helps those who help themselves. With all these things happening one by one recently in sequence I believe he is just telling me to go for it. You know when they say God has plans for everyone.


(Tyler Morgan) #175

Hey everybody,
I started teaching myself how to program about 10 years ago while getting my BS in mining engineering. I worked at mines for a few years after college then ran the alumni community at Udacity for a couple of years and now I’m doing independent consulting back in the mining industry. I’ve done a lot of data work in the past, but the last few years had mostly been doing web work. It’s exciting to be learning something new again and has even reignited old interests in robotics and video games. Hopefully, I will get to apply machine learning to both. Cheers everyone!


(Ng Hui Qin) #176

Hello, I’m Hui from Malaysia, currently working as an iOS Developer in Taiwan. AI is one of my interested field especially the capabilities of creating a lot of interesting applications. I’ve been studying through online courses, such as Coursera (ML by Professor Andrew Ng), Fast.ai (just finish v2 lesson 1), and even apply to be AI Saturdays Ambassador to host AI learning group events. This is my very first decision to jump to AI field after developing apps for 4 years, hope I can learn faster here and move to my target this year.


(Cedric Chee) #177

Hello Hui Qin,

Welcome to the forum :grin: I am another learner from AI Saturdays Singapore. Nice to see more people from Asia here. I can relate to your decision as I also recently take the plunge. It is a big step. Cheers!


(Cynosure) #178

hi everyone!

I started this course end of Oct. I had discovered this course incidently (via meetup.com) on recommendation of some other local students. But this was/is exactly what i needed. I am a developer by profession but did my BS with Maths (but it was almost 17 years ago so i have forgotten almost all of that except the reassurance in my mind that math wont kill me if i have to do/learn it again)

My first attempt to get into ML field was in beginning of 2014 when i read the the book “An Introduction to Statistical Learning with Applications in R” (which is a great free book btw) and then started Andrew ng’s course Machine Learning at Coursera. Unfortunately course was not right for me. it started with some language called Octave and went streight to using gradient descent which i couldnt really understand. So after auditing the course i gave it up … hoping to maybe find another oppertunity or resource some other day. I think the courses at Coursera in meantime have been updated or new launched but my previous bad experience always told me in the back of my mind that it wont work for me. I saw some other online courses which were not free from other websites but having a bad experience i wasnt sure if that money would be worth it or it would work for me.

It took me another 3 years to force myself to start over again, and after revising the above mentioned book was clueless on what to do next. In the hope of building a DL foundation, i had gotten Ian Goodfellows DL and Sheldon Ross’s Intro to Probablity models books but the length of equations in them were intimidating. I couldnt even really start with them. The knowledge i had gained from ISLR book wasnt enough to compete at Kaggle or do any enterprise level project.

so i am very happy to have found this course and I am absolutely loving it. I work full time so have to learn at my own pace. I try to give it around 3-4 hours a day on most days in a week.

Also i really like Jeremy’s teaching style, starting with an intiutive data set of pictures (instead of with some boring decades old crime rate data) and starting from top to bottom with explaination of everything. He is teaching with heart. Thank you Jeremy and Rachel!

I hope to complete Part 1 V2 asap and look forward to the next part.


(Sasikanth) #179

Hello All,

I am from Bengaluru, India working in an IT company. I started learning machine learning from a year ago. I completed Andrew’s machine learning course from Coursera and Udacity ML course. As a logical next step, I always wanted to learn and understand more about neural networks. Its great that fast ai is making “learning neural networks” accessible to everyone. I am excited to start with Part 1(v2) course from fast ai.


(Ahmad Arib) #180

Hi Cedric, I also join AI Saturdays in Jakarta, Indonesia while I’m still in Shanghai, haha. Singapore leading so far in AI field in Southeast Asia, we other SEA guys looking forward to you guys. Please send my regards to Jia Qing whenever you meet him.


(Cedric Chee) #181

Hey Ahmad, thank you for reaching out. Interesting to know from an Indonesian in Shanghai, China. We are all learning from each others. OK and I think you should do that yourself by joining AI Saturdays’s Slack group. :slight_smile:


(Euler Rodrigues de Sousa Faria) #182

Hello All,

I am from Minas Gerais, Brazil and I’m currently working at IBM in Advanced Analytics team. I started the machine learning journey about 07 months ago with an Udemy course ( Machine Learning A-Z ). I’m also pursuing a Master’s degree at University of Campinas on ML field. In the first semester of my Master’s degree I took a Deep Learning class which was given mainly in Pytorch and through this class I could learn and develop some interesting projects. The final project can be found on git hub at this link : ( https://github.com/RenatoBMLR/state-farm-distracted-driver-detection). My partner and I used some interesting techniques, such as transfer learning, ensemble and a coupled of other handy techniques.
I hope to continue this journey in Deep Learning enhancing the knowledge that I’ve acquired so far which I’m certain that still a long way to go and I would like to contribute with the community as well.


(Niyas Mohammed) #183

Hello everyone,

I am Niyas Mohammed, from Trivandrum, India- and I have a great fast.ai story to tell :slight_smile:

When Part 1 was originally released, I was so fascinated by the teaching style of @jeremy and @rachel that I became an evangelist almost overnight. I was on my first job at a well known consultancy, and was one of the few young engineers there desperately trying to understand deep learning. There were no real good resources to start- some were highly technical and mathy, some assumed I knew a lot.

But at the same time we knew that deep learning was definetly going to be the next big thing, and that we did not want to be on the sidelines for this. That’s when I discovered fast.ai.

So a bunch of us worked with our department managers and leads to hold an (unofficial) weekly session on deep learning. And so the Deep Learning Group (DLG) was born. Every week, we’d watch one lesson from fast.ai, work through the week by ourselves and share our experiences at the DLG weekly meeting. This was a fantastic model- people who would otherwise were hesitant to take up online courses showed up and started working together. Everyone loved being able to dip their hands into code and actually see the results. None of this was funded by our company, though. We were paying for all our EC2 compute from our own pockets.

The training offered by our company’s official learning and developement programs are workshops that last one or two days (max) usually. This was a very different, in the sense it was a regular meetup with updates and new things people have tried over the course of a week. After a while, some of the managers became so impressed with our progress and the how such an unofficial meet continued steadily over the course of months, that they decided to procure high end GPU machines for the DLG. That was a fantastic first win for us at the DLG!

And that was not all- seeing the success of this model we built around deep learning, the department started pushing for employees to take up online courses and announced that it would, from that point on, be considered under the company’s “annual mandatory training hours”. Even better, they asked us to form groups just like the DLG for various other domains. So what started out as a small group of ten people meeting up every Tuesday, quickly reshaped how the company thought about learning and development.

This is a huge deal when you think that all this started because of fast.ai and some people in the company who wanted to learn from the very best in the industry

There were great many things that happened since then- I have since moved jobs. I now work at a startup that is working on natural language processing. I started writing blog posts, and took sessions for people getting started in deep learning. Also, I now lead the activities of the Artificial Intelligence and Robotics vertical at the Singularity University Trivandrum chapter.

I get to meet a lot of people at our meetups and share a lot of experiences. And often times, when I am asked what the best place to start learning deep learning is, you can pretty much guess what my response would be :wink:

Great to be here for v2.

I hope we can create some new waves together.

Let’s connect on LinkedIn: https://www.linkedin.com/in/niyasmohammed/