Introductions thread - please say hi!

Part 2 (2017)
62

Hi, I’m Mayeesha. I’m from Bangladesh.

How did you find out about the course? Have you used your knowledge from part 1 on any fun projects yet?

– I found out about the course when I was searching for materials on transfer learning in Youtube. I was doing the capstone project for the Udacity Machine Learning Nanodegree and I had chosen Kaggle’s fishery competitions for that, mostly because the data set was small. Even if I don’t have any particular passion for fishes, the project started getting interesting after I accumulated some knowledge in deep learning and started understanding the overall conversations around DL models. It seemed like even if there’s many good theoretical courses including CS231 for CNN’s, practical courses are super rare, so after I found the youtube lectures I started watching them.

What are you hoping to do with deep learning?

– So far I’ve felt like the entire conversation about deep learning was basically about one topic when I was doing courses at udacity, which is self driving cars. Right now I don’t even know driving, and I am not particularly interested in the self driving cars, except the part that I’d like to ride them if possible to see how it goes. After watching the video on teaching philosophy I was sold on the course.

I wanted to do well in the kaggle competitions that use DL, but I felt those were out of my league. So that was one of my motivations. Right now I’m ranking around 45% in the Fishery one, (I’ve not tried the model with the fully convolutional layers yet though as shown in the lecture), so I feel more confident about my skills in implementing neural networks. My experience in this area is still limited and I’m still trying to understand things, but I want to use deep learning on animation, and basically focus more on story-telling.

Now, there’s not much info available (and I’ve probably not researched well enough yet), how to use Machine learning in the field of animation and I’m still just trying to explore it. To be really honest, I want to know how to make more interesting, entertaining products using ML and specially DL. My general style is basically learning techniques first and thinking about projects later with them. Even in this network visualization project I did before, I learnt about networks and Gephi first and later used it when the opportunity came. It’s a RPG like thing for me, if I know the techniques, I can use them when needed to ‘unlock’ cooler projects. Also I feel like this whole area of ‘using deep learning in creative, artistic project’ area is super super new for me, it’d be a good idea to just explore the field first for a while.

What are you passionate about?

– Reading books, writing, food, cats(I’ve 6 cats), watching anime’s/movies, general social psychology, economics, lots of stuff. I like social networks, human interactions in a broad scale, pretty much random things.

What’s something that not many people know about you?

— I’m not that much anti-social, regardless of my general vibe offline. I’m just introverted. I really like animals and inanimate objects more than human beings. I think I also have a habit of anthropomorphizing things which kind of explains my interest in animation, animals and even robot ethics. I dislike hyper structured environments.

63

Many thanks for sharing this - it’s a very important topic.

2 Likes
64

It is so much fun scrolling through this thread and hearing what everyone is working on! It is great to be back for part II after being part of part I.

@Dario great to hear from classmates in South Africa! My parents are hippies and I spent my childhood moving around Southern Africa, including a year in Nelspruit by the Mozambican border. They are now located in East Africa, in Nairobi, Kenya. I share your passion for using technology to solve problems specific to Africa. Last year and hopefully again this summer, I will be returning to Nairobi to teach a short intro to machine learning course for engineers in the growing tech community. I am a newbie educator but have found it to be an incredible tool to develop empathy and to reinforce my own understanding of material. It has been inspiring to learn from @rachel and @jeremy and has really influenced how I think about the importance of making material feel accessible.

When I do not get to visit family in Nairobi, I work and live in San Francisco. I am part of the data science team at Udemy, an online education company for life long learning. I work on our course recommendation algorithms. I gained so much from part I and with the support of my Director have had the chance to own two projects involving deep learning in Q1.

In the gap between part I and part II of the class I spent most of my free time trying to gain a better intuition of the theory and math underlying deep learning architecture. Understanding what is under the hood has really helped me revisit all our jupyter notebooks in a much richer way. Apologies if this is off topic but in case it is useful the following resources really helped me:

  • by far the best investment I made was in the Deep Learning textbook by Bengio, Goodfellow and Courville. It is so well written, and is great at showcasing the math. It goes through current successful implementation as well as uncertain but promising future developments in the field.
  • Cohlah’s blog is so engaging. He also wrote the paper Concrete Problems in AI safety which I really enjoyed.
  • the deep learning school videos from both Montreal and Stanford. Highly recommend, especially because the target audience is young researchers!
  • This primer on NLP by Yoav Goldberg was fantastic.
  • This may seem strange but one of the best resources I found was a series of AMAs on reddit with practitioners like Hinton, De Freitas, Lecun, Bengio, Google Brain Team, OpenAI. This was important because it helped me understand what questions are being discussed by top researchers today but also paid homage to all the incredible work that has been so far in the field. In a strange way, it helped me place the field within history and also introduced me to some researchers whose work I now follow.

The other exciting project I worked on during the break which I will continue to work on during my free time is applying deep learning to detect illegal deforestation (@brookisme we have to talk!). A few years ago, I started a non-profit in my free time called Delta Analytics that provides free data services to non-profits all over the world. This year we started inviting non-profits to apply where we believe deep learning could be a powerful tool. I am currently co-leading a project with Rainforest Connection to use audio streamed from rainforests in Peru, Ecuador and Brazil to detect chainsaw noises that indicate the presence of illegal deforestation.

It has been a fascinating project and my first introduction to audio processing (so I would love to talk to anyone with experience in the field). The difficulties lie in the fact that:

  • we have very small labelled data population. It is not representative of actual population because some audio was manually created by running a chainsaw in a garden. There is a fascinating paper about this problem here. We are implementing pseudo-labelling to augment our data but due to the original sample composition it is catching very good examples of chainsaws but not less clear cut examples.
  • very high noise to signal ratio, three test sites very different. There are some mosquito and cricket species in particular at the same frequency that can be confusing for even the human who labels the data.
  • how to tell distance of chainsaw from cellphone, so that conservationists are not dealing with a large circumference to search.

I am really excited to potentially apply some of the unsupervised techniques we learn in part II to address our small sample size issue.

Something not many people know about me is that I just started learning archery! My husband has been really into collecting and shooting historical bows for a years but he finally got me to the range at Berkeley. I am totally hooked!

9 Likes
65

Hi everyone,

I am Aman. I work at Visa where I am involved in development of a data platform that helps our business users in deriving behavioral and predictive insights from huge (but structured) datasets (Visa advertising solutions is one of our users, for example).

I came across part 1 of the course via hacker news and loved it. We were working on a recommendation system for bay area residents, and the discussions in lecture 4-5 were very useful in guiding me to the right direction. I’ve done some simple experiments around spend prediction (a.k.a. time series forecasting) using a multilayered perceptron and the results are encouraging despite the small dataset (size of the dataset shouldn’t stop one from exploring deep learning solutions was one of the good points taught during part I of the course!).

Before joining Visa, I was a master’s student at IIT Bombay, India where I worked on NLP (specifically, on the problem of numerical relation extraction from unstructured data).

I am very excited for the second part of the course. I’ll especially look out for NLP related applications and for techniques that can help in scaling deep learning solutions to large structured datasets.

For more details on my background and experience, please see https://www.linkedin.com/in/amnmadaan/.

P.S. I missed the first lecture because of prior work related commitments, but will be attending in person from the second lecture. If you are looking for a team member and my background seems relevant, please let me know.

Thanks!

3 Likes
66

Hi Constantin!

Can you elaborate a bit about the advantages of using bcolz? I heard great things about it, but i didn’t see it’s benefit besides being a fancy API for saving numpy arrays. what am I missing?
BTW, I’m also doing the cancer challenge. will love to share some ideas :slight_smile:

67

@Even

Not sure about generating the whole book, but this might be a good dataset for generating covers.

http://dlp.lib.miamioh.edu/picturebook/

Has abstracts, keywords, and cover images for 5700 children’s books. Looks like you’d need to do a bit of scraping to get it all though.

68

Hi @davecg,
From my point of view and according to my experience with tensorflow, I think it is possible to generate the whole book with respect to the input choice (the output will depend on what we fill in the network input).
We could discuss deeper later with all things we will learn here.
Thanks.

1 Like
69

@davecg I’m not sure. I’ll ask them. Jaishree is head of Radiology at the Nelson Mandela Children’s hospital at the moment.

70

@jeremy that would be super awesome :slight_smile: will do!

71

@sara.hooker13 Kudos for teaching ML in Nairobi! I’ve been looking for a way to set up some sort of academy for data science in sub saharan Africa - if you’re interested maybe we can set something up together :slight_smile: I’ve also been looking into using Machine Learning to create education at scale in Africa (and elsewhere) by building a smart tutor using a chatbot type interface. The idea is that one person with knowledge on a subject does a Q&A session with thousands of people via the web, but the questions are filtered and grouped by the chatbot as an intermediary so that many people’s questions can be answered at once. Not sure how to start on this really (I’ve played around with chatbot tools offered by IBM and Microsoft, but they seem too restrictive), but I am excited to see what we’ll learn about chatbots in this course!

2 Likes
72

Hello! After avoiding this thread for a while, here it is:
I’m Slav, entrepreneur/hacker based in Sofia, Bulgaria. My passion for tech that has an impact has led me to a few different ventures, from enabling people to plant trees to working with large brands.
I heard about part 1 of the course being the best thing for practical machine learning, but decided to focus on kaggle first. Well, kaggle competitions turned out much harder than I though so I came back and instantly fell in love with the way part 1 presented the material and finished it just in time for part 2.
Many, many thanks to Jeremy and Rachel for doing this!

I’m hoping to apply DL to EEG brain waves, to better understand intent/feeling/thoughts. The hardware that I have right now is crude, but it’s a start. Computer vision is really interesting for me, as well, as it can be applied to various tasks in autonomous drones.

What many people don’t know about me is that I catalogue and count all my possessions and try to reduce them to as few as possible (i.e minimalism).

Thanks for reading!

5 Likes
73

Hi everyone! I’m getting started a bit late and hoping to catch up with you soon:)

I’m Ella, biologist and a software engineer. Icelandic mother of three boys and live currently in Stockholm, Sweden. I’ve done all sorts of things in my carrier like many of you here, started off in protein-protein interaction studies on BRCA2 breast cancer gene, moved to working in the airline industry mining all sorts of data for 9 years before diving again into the gene-pool and running all sorts of algorithms on raw sequencing data of the human genome in preparation for scientists. Today I’m working for Qlik, mostly in research around data engineering at the moment.

I’m very happy to get the opportunity to participate in part 2.

2 Likes
74

Hi everyone, my name is Xin WANG from China. I work for a Dutch Coating company as IT manager based in Shanghai. Currently I am leading a small group trying to introduce Machine Learning technology to our company.
•How did you find out about the course? - I found the part1 on Youtube. The content is really helpful to one of my project (identify our new product from a picture of our local stores). We use VGG16 and ResNet as basic technology.
•What are you hoping to do with deep learning? - We are also planning to develop customized ChatBot for our business operations in several area
•What are you passionate about? - Discovering more and more successful business case of applying ML technology in a traditional insdutry is very exciting
•**What’s something that not many people know about you?..**Besides technology, I also like art and history. I practice Chinese calligraphy everyday

We also setup learning group about this fantastic course in WeChat (a chinese version Twitter) with around hundreds of members

1 Like
75

Hello everyone I’m Ravi and I’m from India.I work as a data analyst in a software firm.I found out about kaggle about a year ago and from then I have been participating in them. I heard about this course in kaggle forums.I have thoroughly enjoyed part-1.I’m amazed at how simple high-school math can be capable of producing cancer-detecting algorithms.

I am hoping to find better medical image analysis algorithms with deep learning.Just like how image-net revolutionized object-detection and image-classification, in long term I want to create a medical-image database and let people develop algorithms which can revolutionize medical imaging.
I am still a beginner in programming, deep learning etc and hope to contribute to open-source software that I use.

Thing that many people don’t know about me is that I read/watch a lot of manga/animes.