We are excited to offer a limited number of travel grants to students and postdocs to attend CVPR and Multi-Agent Behavior workshop. Each award will cover at least $500 towards travel and meeting attendance costs (larger grants may be available to those travelling longer distances.)
You will get a chance to meet a panel of experts from various disciplines and discuss the key goals of multi-agent behavior research. CVPR is the premier annual computer vision & pattern recognition event regarded as one of the most important conferences in its field.
Apply for travel grants by submitting an extended abstract or paper to Round 2, and we will review travel grant applications submitted between now and May 27th on a rolling basis.
Round 2 Submission deadline: May 27th, midnight AoE
Submit at: Conference Management Toolkit - Login
Travel grants with submissions reviewed on a rolling basis.
Final Decisions: June 3rd
Camera-ready Deadline: June 10th
Just a quick reminder, we’ve extended Round-2 deadline by one and a half months after receiving your feedback that the current timeline was too short to work with video data.
[Name], now with the new deadline of 3rd July, 2022, we look forward to seeing your unique submission.
Don’t miss the chance to present your work at CVPR workshop by participating in Round-2.
Round-1 saw 224 participants making 700+ submissions. We thank you all for your participation. Here are the winners of Round-1 of Multi-Agent Behavior Challenge 2022.
"I used a Bert style encoder, treating handcrafted features as the tokens and performing masked language modelling. Initially, every frame of key points is converted into a large number of handcrafted features representing angles, distances and velocities between body parts within an animal as well as features from each animal to all others. For the flies, I had 2222 features and 456 for the mice (but would use more in the future, especially for the mice) and these features are all normalised.
I’ll refer to the neural network as having a head, body and tail where the head and tail act on the single frame level and the body acts on the sequence level. The head is a two-layer fully connected network that reduces the input features down to the target dimension size (256 or 128). The body does the “language modelling” part - input is 512 partially masked tokens (masked before the head) and output is a sequence of the same shape (batch_size, seq_length, 128 or 256) but now hopefully includes higher-level sequence features. I used huggingface’s perceiver model for this - Perceiver.
The tail is a single linear layer and can be thought of in two parts, original unmasked features reconstruction and predicting any known labels so for example that would be of size 458 for the mice. The loss function I was using was mean square error for reconstructing the features and cross-entropy loss for the (non-nan) labels where the mean square error loss was weighted approximately 10 times more." Stayed tuned for a more in-depth breakdown of his solution.
In the new round, you’ll be given two sets of raw overhead videos and tracking data. As in Round 1, we ask you to submit a frame-by-frame representation of the dataset. We hope this video dataset will inspire you to try new ideas and see how much incorporation of information from video improves your ability to represent animal behaviors! Explore the sub-tasks to know more
The end goal remains the same, create a representation that captures behavior and generalizes well in any downstream task.
As mentioned in the Round-2 announcement for Multi-Agent Behavior Challenge participants who make a submission in round-2 can win up to $400 AWS credits.
Keep reading to see how you can claim AWS credits for yourself or for your team
- You (or your team) must have received scores higher than the below-mentioned scores in any of the 2 tasks.
- On the Ant & Beetle Video Data task, you must beat a mean F1 score of 0.606. Here’s the baseline for the Ant & Beetle Task, which achieves a mean F1 score of 0.591.
- On the Mouse Triplet Video Data task, you must beat a mean F1 score of 0.306. Here’s the baseline for the Mouse Triplet Task, which achieves a mean F1 score of 0.292.
- Each participant (or team) can claim a $200 credit code for each target beaten (up to $400 total.)
- We will be sharing 50 X $200 = $10,000 worth of AWS Credits. The distribution will happen on a first come first serve basis and will cease once all the 50 codes (each for $200) are distributed.
- For teams claiming the code, please nominate one person to represent. The code will be mailed to that team member.
How to redeem?
- Please share the following as a reply to this thread to receive your code.
Team name (if relevant):
Submission id (the one that beats baseline as defined above):
How much did you improve over the relevant baseline score?:
A brief intro about you (Us and the participants would love to know what brought you to this challenge):
We will be sending out the AWS credits every week after verifying the details. After receiving your code, you can go to this website and claim your credits.
If you haven’t gone through the baselines, check them out here
All the best!
The new round of Multi-Agent Behavior Challenge 2022 is live!
As the Round 1 of this very unique & exciting challenge comes to an end, we are happy to share the new prizes, dataset and deadline for Round 2!
Everything! In this round you’ll be given a dataset of raw overhead videos and tracking data. Rather than being asked to detect a specific behavior of interest, we ask you to submit a frame-by-frame representation of the dataset. This video dataset is unique will inspire you to try new ideas that they could not on just the tracking data. Explore the sub-tasks to know more
The end goal remains the same, create a representation that captures behavior and generalizes well in any downstream task.
$6000 USD cash prize pool for the two subtasks.
Round-2 runs till May 20 th, 2022, 23:59 UTC.
Claim upto $400 AWS credits in Round-2
These notebooks will help you perform initial data exploration and a basic embedding using a vision model.
There is a $6000 cash prize pool for the new round. For top positions on the leaderboard, each task carries a $3000 cash prize pool.
1st on the leaderboard: $1500 USD
2nd on the leaderboard: $1000 USD
3rd on the leaderboard: $500 USD
Bonus: Teams that beat the advanced baseline can stand to win $400 AWS credits. Details on the score to beat and the advanced baseline will be announced soon. !
Top participants and teams will be invited to speak at CVPR 2022 Workshop. You will get a chance to meet a panel of experts from a variety of disciplines and discuss the key goals of multi-agent behavior research. CVPR is the premier annual computer vision & pattern recognition event regarded as one of the most important conferences in its field.
Check out the latest updates, datasets and baselines for Multi-Agent Behaviour Challenge.
Watch the Town Hall Here
What does this town hall include?
Challenge organiser, Dr Ann Kennedy from Northwestern University, presented the overview of round 1 of the Multi-Agent Behavior Challenge. She also discussed the embedding evaluations.
Andrew Ulmer from Kennedy Lab @ Northwestern University presented the tVAE baselines.
Jennifer Sun from Caltech presented the details of Round-2 of the challenge. She also previewed the new dataset.
Participate Now: AIcrowd | Multi Agent Behavior Challenge 2022 | Challenges
You are invited to join Multi-Agent Behavior Challenge 2022 Town Hall event on 03 April Sunday, 3 PM UTC.
The town hall will feature challenge organisers, Dr Ann Kennedy & Jennifer Sun, from Northwestern University & Caltech. They will discuss details about the new round, preview the dataset and present new baselines.
Multi-Agent Behavior Challenge 2022 Town Hall
03 April Sunday, 3 PM UTC
Since Round-1 of Multi-Agent Behavior Challenge 2022 is extended to 11th April, this event will help you improve your solutions. Round 2 of the challenge will also be announced soon. Get details about the upcoming datasets and evaluation metrics.
Share your feedback and experience from Round 1 with challenge organisers.
Make the most of the new deadline by getting the latest challenge updates, baseline overview and preview of the new dataset.
Challenge organiser, Dr Ann Kennedy from Northwestern University, will present the overview of round 1 of the Multi-Agent Behavior Challenge. She will also discuss the embedding evaluations.
Jennifer Sun from Caltech will present the details of Round-2 of the challenge. She will also preview the new dataset.
Andrew Ulmer from Kennedy Lab @ Northwestern University will present the tVAE baselines.
If you want to get the latest updates, improve your solution or discuss new ideas with other participants, don’t miss this Town Hall event this Sunday.
Got a burning question? Get it answered by challenge organisers and the AIcrowd research team during the event. Drop your questions and query before the event on this post.
Meet your fellow challenge participants IRL. This is your chance to discuss new ideas and approaches with other AIcrowd community members.
This repository contains jupyter notebooks that implement a trajectory variational autoencoder (tVAE) baseline model for embedding the mouse and fly trajectory datasets for the 2022 Multi-Agent Behavior Challenge.
Performance of these notebooks is as follows:
|Dataset||Mean F1||Mean MSE||task F1 1||task F1 2||task F1 3||task F1 4||task F1 5|
Where the “task F1” values are F1 scores on specific sample evaluation tasks. Note, while this baseline is outperforming PCA for the mice, it actually does significantly worse than PCA for the flies! (Why? Who knows!)
Clone this repository and open train_mouse_tvae.ipynb or train_fly_tvae.ipynb in a jupyter notebook session. Follow notebook instructions to make your own submission, then play with the model architecture and parameters inside the
tvae directory to see if you are able to improve performance!
Note, you’ll need to download the mouse and/or fly datasets into the provided
fly_data directories to use this code.
Got questions about the new baseline? Drop them in the comments below
Hope you are enjoying the Multi-Agent Behavior Challenge 2022. This is a very exciting problem and we are glad to have you on board! We understand that this unique challenge might require some extra time so we are extending the deadline for Round 1.
There is a $6000 USD cash prize pool for the two subtasks.
The new deadline for the first round is April 11th, 2022, 12:00 UTC.
Detailed individual task scores are now available on the submissions page. This will help you identify & understand which ideas improved different scores.
Have any more questions? Comment below and we’ll answer.
You can always post your questions and issues on the challenge forum
At AIcrowd, we have a very creative team working on new AI puzzles for you to solve. They set the problem, brew the dataset and create solution notebooks.
What kind of puzzles do you want to solve next?
Suggest puzzle ideas or new topics we should explore. Comment on the types of puzzles you’ll like to see. Want to collaborate with us to create a unique AI puzzle? Drop an email at firstname.lastname@example.org.
AIcrowd Blitz is for you. So help us make it better by creating problems you want to solve!
We have built Blitz through extensive user interviews and in-depth surveys of existing participants. We want AIcrowd Blitz to be a comprehensive AI companion. Blitz is a place where you can learn, solve and show your AI skillset.
We want to make Blitz better for you. We want to hear your unfiltered, honest thoughts on Blitz. You can also answer these questions (either through comments below or by email at email@example.com).
- What is one thing you love about Blitz?
- What is one thing you immediately wish to change about Blitz?
- How would you rate your Blitz experience so far? (Out of 5, where 5 is excellent and 1 is poor)
- This can be the user experience, quality of puzzles and solutions and overall ease of subscription and access to Blitz.
- How would you rate Blitz AI puzzles? (Out of 5, where 5 is excellent and 1 is poor)
- This can be about the puzzle’s difficulty, the quality of the dataset or the puzzle content.
- How would you rate Blitz AI solutions? (Out of 5, where 5 is excellent and 1 is poor)
- This can be based on ease of understanding, explanation quality, and implementation ease.
- How easy was it to deploy your first ML App? (Out of 5 where 5 is very easy and 1 is very hard)
- This can be based on the ease of implementation and sharing of the app.
- Would you recommend Blitz to a friend? If not, why?
- What is one item we should improve?
We have tried to incorporate features that will accelerate your AI learning. We have also polished & perfected the things you love about Blitz.
Like you, we love learning AI & solving new problems to build our skills. So we want to hear your thoughts on AIcrowd Blitz. Please share your honest feedback to help us provide the best AI platform to you. We are always available on @AIcrowdHQ on Twitter and firstname.lastname@example.org.
For the last 1.5 years, we have been developing unique AI puzzles and sharing them with you every month. The next phase of the evolution of AI Blitz is here!
AIcrowd Blitz is a comprehensive AI Learning companion with unique AI puzzles, expert solutions, your own AI community & most excitingly, an impressive ML portfolio.
AI Puzzles You Love + Easy to understand Expert Solutions
Blitz boasts a library of hand-crafted AI puzzles designed around home-brewed datasets on real-world problems. Signup and get these unique AI puzzles delivered straight to your inbox.
Learn As You Go
Unlock solutions and resources created by experts for beginners. Learn different approaches to solve AI puzzles, understand the theory and gain hands-on experience.
Build & Share Your ML Portfolio
Create AI Solutions For The World To See
Don’t just sharpen your skills; show them to the world! Through Blitz ML Apps, make your solutions accessible to all. Host live ML Apps for anyone to use in real-time. With Blitz, you can easily build a stellar ML Portfolio.
Meet Your AI Tribe
We know learning is hard, so we never want you to feel alone in your AI journey. Through Blitz, meet like-minded AI enthusiasts that will inspire & help you. Get exclusive access to live problem-solving events and 24*7 community support.
Like you, we love learning AI & solving new problems to build our skills. So we want to hear your thoughts on AIcrowd Blitz. Please share your honest feedback to help us provide the best AI platform to you.
As the Round-2 of Data Purchasing Challenge 2022 comes to an end, let’s shine a spotlight on the winners.
Community Contribution prize winners are being finalised and will be announced soon, stay tuned.
Data Purchasing Challenge Round 2 is going strong To help you make the most of the challenge, we have released a new baseline containing fast heuristic implementations of some simple ideas.
If these notebooks and explainers help you, don’t forget to the notebook and leave a comment.
Explainability: How does your model actually learn? Another notebook by aorhan, explain how given a sample you can identify what features contribute to the decision. And how you can improve your model. Read the notebook over here.
Additional Resources from Challenge Organisers and Top Participants: We recently hosted a live Town Hall event where the organisers and participants shared their ideas. You can find the recording & resource compilation over here.
Solve AI Puzzles, Build ML Apps.
Grow Your Skills With The Community. Show Your Work To The World.
Comment your questions in the thread below or email email@example.com
Get your queries on waitlist, launch, features & more answered quickly.
Drop your questions in the comments below