BIRDS
Beginners in Research-Driven Studies
Channel: #beginners-in-research
Hello there, fellow learner! Glad you found our channel. You might be wondering, what is it that we do here? We are a group of people obsessed with learning new things and sharing our insights. But more importantly, we're a group of people dedicated to fostering a passion for research among beginners. We're BIRDS(Beginners in Research Driven Studies) of C4AI.
Whether you are new to research or looking to refine your skills, we are here to support you in your journey. We believe that by working together, we can all grow and learn more effectively. Welcome to our Channel!
Co-leads:
Herumb[@krypticmouse on Discord, @krypticmouse on Twitter]: Hi, Herumb this side! I'm just a daily life shinobi obsessed with mangas and machine learning. I love learning and I like to help others learn as well. I like the raised hand emoji a lot. If you think I can help you out in someway feel free to reach out on Discord or tag me on channel. Whatever you are comfortable with. See you around!
Akanksha - @Ash7#2421 on Discord, @akankshanc on Twitter
Caroline Shamiso Chitongo - @c.s1693 on Discord - linkedin.com/in/caroline-shamiso-chitongo-7724381a8 on LinkedIn
Reza - @rzsgrt#9099 on Discord, @rzsgrt on Twitter
Goals
To introduce beginners to the fascinating world of research.
To cultivate a supportive community where questions, ideas, and discussions are always welcome.
To share resources and opportunities that can help our members grow as researchers.
To organize events, such as workshops or seminars, to educate and inspire our members.
Logistics:
We communicate primarily through this Discord channel on channel #beginners-in-research.
Occurrences: weekly, Friday at 1:00 pm ET: https://meet.google.com/zsr-dbvv-ofh
Summary Banks
Summary Bank #1: Conformal Prediction
Hi BIRDS! This is the first summary bank, took a bit longer than we expected mainly because we can’t post long articles on Discord and short ones were well…not too good. Frankly speaking if we did want to do this it would be because it helps you understand a vague idea of a topic or domain in one go, rather than resource hoarding. So what is this bank about? Well, this one is about Conformal Prediction. It’s not really my favorite topic to talk on to be frank, nothing against it it’s just not something I use in daily life but it’s a really cool topic to understand!
What is Conformal Prediction?
Conformal Prediction comes under the topic of Uncertainity Quantification, which is quantification of uncertainity. That doesn’t seem very helpful does it? Let me elaborate, in the most simplistic setting of Deep Learning problem we are trying to find an answer. This answer can be predicting the breed of dog, type of object or maybe the next word too. Given some inputs to the model, we get this answer and that’s usually all that we care about. This type of setting is called Point Prediction where point is the answer we get from the model. Uncertainity Quantification can provide us a way to find how certain or confident the model is about these predictions.
Conformal Prediction also provides a way to generate a prediction set for any model. To put it simply:-
Conformal prediction uses an already trained model to estimate “all” predictions that a model can make for the given input with high confidence.
Suppose your model predicts a breed of a dog, with conformal prediction you can get the possible prediction set for that image. So given an image instead of breed as an answer you’ll get a list of breeds possible for that image. The catch is that Conformal Prediction provides us a guarentee that this prediction set fall under a high probablity range that you set. This guarentee is called Coverage.
How do I do it?
So, we know a bit about conformal prediction but how do we do it, say…in code. Well let’s talk about the easier way first, its as easy as cooking instant noodles. You have the tools(libraries) just follow the instructions and you should be good. So, you essentially have 2 steps:-
Training Step: Training a model on the point prediction task on a dataset.
Calibration Step: Estimating the Confidence interval/Prediction set of the prediction from the model over a held out IID dataset for this task.
During the calibration step, we do the following:-
Decide a scoring function and get the score of the correct class. This score could be absolute difference or anything to be fair. So you feed input to the net and let the softmax vector from that vector find the softmax value of the correct label and find its score. Note this softmax value of correct class that the model predicted may or may not be highest.
Repeat the above for all the points in the calibration set and get the “score list”. Take ~10% quantile of these scores. With this you get a score(let’s call it q) such that 90% points/scores in score list have a true class score more than q.
Now during prediction, when you get softmax score for the outputs all the score that are more than q form our prediction set.
I’m in where do I go?
Cool, really glad that you liked conformal prediction. Darn, now I really wish it was a blog post but I have really cool resources you can refer to learn about it!
A Tutorial on Conformal Prediction(Playlist, basis of the section 2 example): https://www.youtube.com/watch?v=nql000Lu_iE&list=PLBa0oe-LYIHa68NOJbMxDTMMjT8Is4WkI&ab_channel=AnastasiosNikolasAngelopoulos
A Gentle Introduction to Conformal Prediction and Distribution-Free Uncertainty Quantification: https://arxiv.org/abs/2107.07511
Introduction To Conformal Prediction With Python: https://christophmolnar.com/books/conformal-prediction/
Conformal Prediction in 2020: https://www.youtube.com/watch?v=61tpigfLHso
Resource Stack: https://github.com/valeman/awesome-conformal-prediction
Summary Bank #2: Mixture of Experts
Coming Soon...
CUDA Programming Mini-Cohort
May 10, 2024
May 3, 2024
April 26, 2024
April 19, 2024
Introduction to Concurrency - April 13, 2024
Mini Cohort Overview Session - April 5, 2024
Recent Presentations
March 15, 2024
March 8, 2024
February 23, 2024
Journey into AI with Lewis Tunstall
January 19, 2024
January 12, 2024
January 5, 2024
Unveiling the Journey with Edward Hu - Insights into LoRA, μTransfer, and the Art of Reasoning, Plus Valuable Advice for Beginners and More!Link to his paper: https://arxiv.org/abs/2106.09685
Tutorial on deepspeed. Topics covered:
- How people can use deepspeed via different libraries like lightning, hugging face transformers etc. to training there models much faster.
We discuss: "ZeRO: Memory Optimizations Toward Training Trillion Parameter Models"
Link to paper: https://arxiv.org/abs/1910.02054
Katie Matton - "Unveiling her amazing research journey at the crossroads of machine learning, affective computing, and behavioral health"
Introduction to Federated Learning
Intro to the latest sprint - 'Circuits - Zoom in' paper by Chris Olah et al.
Herumb presents of LangChain
high level implementation of ViT
Paper replication sprint overview: An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale (ViT)
Nidification: Tokenization in NLP
"Parameter Efficient Fine-Tuning" - implementation showcase. See the work of the BIRDS on their first paper implementation sprint.
Hailey (@hails) - 'Flying lessons with ex-Birds' - guest speakers walk us through their research journeys
Tips and Tricks for Doing Good Research with Sara Hooker
Materials from all past sessions