This year during Hactoberfest we decided to run a month-long study group in the form a WhatsApp group. In which we also did few AMAs.
The following is a trimmed and slightly reordered transcript of the AMA with Ali. If you want the full chatlog you can access it here.
Ali is the organizer of Tensorflow India Community. He’s also part of many community organizing groups and likes to interact with engineering students at large. He speak a Lot :p ). More about him here
AMA was conducted on 22nd October 2019 from 8-9PM.
- A: Ali
- P(n): Participants
P1: Where do you see tensorflow reaching within 2-3 years down the line?
A: With the release of TensorFlow 2.0 it makes a lot of sense that Google is investing heavily and this time keeping community-centric. Also earlier the Usergroup’s were not supported by Google now they have started supporting them. It has a list of features and realizes planned. You can follow the GitHub for the latest updates or be the part of TensorFlow Google Group(open for all) to see the latest developments.
P2: What are the best ways for one to learn tensorflow in India?
A: There are pretty good materials online. I highly recommend a series of 6 Codelabs available on Google Codelabs I’ll share the URL here itself in some time. Also documentation of TF 2.0 is very easy to understand and one can start from it. After Tensorflow.world more new resources would be available. For a production basis most of the companies Prefer Tensorflow and for a research basis most of them use PyTorch. If you are looking as a Developer it’s worth taking the effort to master it.
- LAB 1: https://codelabs.developers.google.com/codelabs/tensorflow-lab1-helloworld/#0
- LAB 2: https://codelabs.developers.google.com/codelabs/tensorflow-lab2-computervision/
- LAB 3: https://codelabs.developers.google.com/codelabs/tensorflow-lab3-convolutions/
- LAB 4: https://codelabs.developers.google.com/codelabs/tensorflow-lab4-cnns/
- LAB 5: https://codelabs.developers.google.com/codelabs/tensorflow-lab5-compleximages/
- LAB 6: https://codelabs.developers.google.com/codelabs/tensorflow-lab6-largecnns/
P3: Hi, I was wondering if you could shed some light on how do you go about introducing Tensorflow to people who have programming knowledge but have little to no prior information on Machine Learning? Do you have a hackathon based approach, where the focus is on learning by doing/completing projects? Or is it more on the theoretical knowledge at first? I am assuming the aforementioned participants here do have some form of mathematical knowledge in regards to basic statistics and the ilk.
A: I have Trained a lot of students and being a student I feel we get bored out in Theory. The approach I follow is a very practical approach with a little Important Theory part. (Very essential for interviews) By making things they to do mistakes and then when you figure a way to solve the error you learn a lot. End-To-End projects are very important. I have tried the Hackathon based approach if you keep the audience small works like a charm. Maths always comes later. Is it important? Yes very much. But messing with it earlier makes people go away from ML itself.
P3: AH! Thank you so much! 😀 This helps me a lot. I do have a followup question, do you aim to complete a basic project with every session like this? Since you stated that you place a higher priority on End-to-End projects? Is this to show the entire process of the development cycle to beginners? And if so, how do you keep it from overwhelming the participants?
A: As they would be surprised in the near future to see the whole process and then blame the trainer as he/she showed a half picture(usually 20-30% is covered in such training and majorly focused on training/building the model). Yeah there can be an exception when you have an audience for a specific topic or use case
P4: What are the future aspects of tensor flow in today’s market?
A: If you see the current scenario it’s more driving to Community first and it’s quite visible with TF 2.0 realize They have invested heavily in Tensorflow low-level API for research as well that allows you to build onto the internals of TesnroFlow without having to rebuild it
P5: Do you see tensorflow becoming an Industry-standard soon? What steps are being taken to encourage widespread adoption of tensorflow?
A: The first step is Tensorflow.world. As you can well see it’s a major conference happening just around TF. Also you might have seen TF Roadshows happening around the Globe
P6: Do you have any plans for Guwahati to conduct TensorFlow. If you are conducting kindly let us know
A: Recently there was an event TesnroFlow All Around happened in Guwahati. I plan to visit all chapters in-person 🙌 by mid 2020. We have 13 active chapters across the Country.
P7: Can you please share few resources of TF Lite for flutter?
A: I haven’t worked on flutter at large but for Tensorflow for lite there are good resources available online one of my favorite is Tensorflow for poet’s lab.
P7: What do you think like why one should use the tf over pytorch?
A: I don’t give priority to TF over Pytorch or vice-versa both have many advantages and disadvantages. Pytorch is good for research-based work but when I say taking your app to Market TesnroFlow is the king there.
P7: What are your pro tips for the developers who want to become a speaker? Also, you may share with us any awesome TF project you have worked on!!
A: Better speaker means better practice. I am good now just because I have done so many talks. Dev’s are mostly nerds and Talks needs Engagement so practice with the small audience then gradually increase it. Yeah I have awesome projects on TF but I haven’t put them up in GitHub I show them live (As I am still writing Papers for them)
P8: Can you share some thoughts on the current light on the current community scene and what else can we do to make things more enriching and interesting for beginners. Also, I’ve got to know you’re into crowdsource initiative also Share some light on what is it and if it helps in solving the data problem
A: Community is growing and now we also have support coming in from the Google Team. We will be conducting events/Hackathon and many more activities for beginners! Always encourage them. Don’t just redirect everyone to Andrew’s course! Understand what they want to do then suggest them resource’s. Make sure you find the right one and if you can’t find a GDE for ML who could help. But I have seen it as a trend. I find people doing it all the time😅.
With my network in India, we were able to lay the base for Crowdsourced community in India. Have you used Google Assistant in the native language? How much improved it is as compared to a year back? We have crowdsourced the data for them. Also, the project InclusiveImages where my better half MD. Nawaz presented paper at NIPS(now Neural IPS) on how we collected ~5 million labeled images within 30 days.
P9: When you talk about community crowdsourcing is it related to fundraising?
A: I am no longer part of the Google Crowdsource Team but continue to do Crowdsource I’ll be speaking at TensorFlow.world on how We remove Bias in Machine Learning using Crowdsource in Santa Clara on Oct 29th https://www.youtube.com/watch?v=wwZKOmoNxMo Might help you understand it (feel free to skip first 5 mins after that main talk starts )