2024 AI Competition for UK University Students
SNN-Based Applications
We are delighted to introduce SECQAI’s SNN-based Application Competition, an exciting avenue for university students to demonstrate their prowess in constructing applications based on Spiking Neural Networks (SNNs). At SECQAI, our commitment to fostering innovation in artificial intelligence drives this competition, offering a unique platform for aspiring minds to delve into the vast potential of SNNs.
The focal point of the competition revolves around crafting an SNN-based application. 50% of your application must be made up of an SNN, however you may choose to combine this with other architectures to create your application (e.g. a Convolutional Neural Network could feed into your SNN). Your SNN solution should be at least three layers, each hosting up to 1024 neurons of the Leaky Integrate-and-Fire (LIF) type. Participants are encouraged to select applications aligned with their interests, and collaboration is permitted, allowing teams of up to five members.
Submissions to the SECQAI competition undergo evaluation based on applicability and real-life usefulness metrics, along with computational resource usage metrics to gauge solution efficiency. The comprehensive assessment includes considerations such as source code, a succinct write-up, innovativeness, commercial relevance, and more. A SECQAI committee will holistically review submissions for a monetary prize (details outlined below), emphasizing the competition's dedication to recognizing outstanding contributions in the realm of SNN-based applications.
You can register your team for the competition at the following URL: Team Registration is now closed
But, before you do, don’t forget to read all the details below!
Competition Details
1. Important Dates:
Registration Deadline: 31/01/2024
Submission Deadline: 31/03/2024
Winner Announcement: 15/04/2024
2. Prizes
Prizes will be awarded to the top 3 teams; they will be split equally between team members.
1st Place: £1000
2nd Place: £750
3rd Place: £500
3. Competition Submission
Your competition submission will require the following:
Source code for your SNN application: To ensure a smooth validation process for your submission, it is imperative that you provide well-organized and comprehensible source code. The code should be easy to follow and execute seamlessly on our end.
Report discussing your application: Concise, yet comprehensive, addressing key technical points, including any technical data sets you have used. This report should offer insights into the underlying architecture, design choices, and any significant algorithms employed in your project. Include relevant details that can aid in understanding the functionality and purpose of your application.
4. Detailed submission requirements
a. Your application solution is expected to adhere to the following specifications:
At least 50% of your application must be an SNN: While not required, you may choose to combine your SNN design with other approaches. However, at least 50% of what you build must be an SNN.
Target Three Layers, but this is flexible: Three layers will ensure simplicity and efficiency in the network’s architecture, however we understand that your application may need more or less layers.
1024 LIF Neurons per Layer: Each layer within the Spiking Neural Network should not exceed 1024 Leaky Integrate-and-Fire (LIF) neurons. This specific limit defines the size of individual layers, promoting scalability and resource efficiency.
Implemented in Python: The entirety of the project, encompassing both the training and testing phases, must be developed using the Python programming language. This requirement ensures a standardized and accessible codebase.
Datasets: Participants may use any opensource dataset of their choice that works for their developed application. However, participants should clearly identify which dataset is used and explain what parts of the dataset is used and how data were split between training and testing subsets.
b. Source code requirement
Challenge participants must provide the source code used in the creation of their solution (model definition, final trained model, training scripts, inference scripts, etc.) with MIT or BSD3 license.
We should be able to run your solution and verify your results
By submitting your solution, you warrant that it:
Does not contain any content that infringes on any third-party Intellectual Property (IP) rights, and that you own or otherwise have all rights necessary for the submission including any and all IP rights
Does not disclose any information which would constitute a violation of a confidentiality obligation.
Does not contain any viruses, worms, spy ware, or other components or instructions that are malicious, deceptive, or designed to limit or harm the functionality of a computer.
c. Report
A concise report, limited to a maximum of 5 pages, detailing the ideation and development process of the solution. This documentation should encompass the following key aspects:
Overview of your Application:
Clarify the purpose and functionality of the application.
Provide a high-level understanding of its intended use and impact
SNN Architecture Details:
Elaborate on the design and structure of the Spiking Neural Network (SNN).
Offer insights into the key architectural choices made during the development process.
Code Documentation:
Supply brief, yet comprehensive, instructions on how to train the model and execute test set inference.
Include any essential prerequisites and dependencies necessary for successful implementation.
Instructions for Running and Testing:
Clearly articulate step-by-step instructions for running and testing the application.
Highlight any specific configurations or parameters crucial for optimal performance.
Additional Considerations:
Reflect on what aspects of the solution proved effective.
Discuss challenges encountered, detailing what did not work as anticipated.
Justify the selection of specific strategies over alternative approaches.
For your report, please use a single-column Word document or Latex template with 1-inch margins, single-spacing, reasonable font size (11pt or 12pt; default font like Times New Roman), and up to five A4 pages. Please submit a PDF.
5. Evaluation of your submission
A committee of SECQAI team members will evaluate submitted solutions to decide the winners. The committee will make a holistic evaluation covering applicability of the solution, complexity, computational resource usage (hardware efficiency), solution write-up quality, innovativeness, and commercial relevance.
6. How to participate
To join the competition, follow the registration instructions outlined below. The challenge timeline overview in (1) above has been provided for your reference. Upon successful registration, you will receive timely updates regarding the various phases of the challenges, ensuring you stay informed and engaged throughout the competition.
Submit your registration via the form at https://www.secqai.com/ai-competition-registration by the deadline
Note, registrations will only be accepted if they are received with valid UK University student email addresses
The registration form includes the following fields:
The name of your team (be creative!)
The university you attend
The details for each team member, up to a maximum of 5 (Full name, university contact email, degree being undertaken, year of study, any previous degrees held and the awarding university)
7. Helpful resources
SNNs represent robust tools capable of addressing complex challenges, including classification, decoding, regression, recognition, and more. If you just look at classification as an example (aligning nicely to hardware efficient SNNs), there are plenty of different applications:
Sound classification:
Musical instruments
Spoken digits/commands
Speaker identification
Biomedical signals classification:
Hand gestures (EMG)
Normal/abnormal heart activities (ECG or PCG)
Etc.
You have the complete freedom to select the application that you want, ensuring that your choice aligns with your interests and objectives, keeping in mind the evaluation criteria we mentionedin (5).
You can have a look at the following resources to give you a head start on the topic area.
We recommend using Colab to develop and test your implementation.
Tutorials on basic Python and LIF neuron models: LIF-1 and LIF-2.
Dataset search engines: Google Dataset Search and Kaggle.
Further readings on SNNs and their applications: paper1 and paper2.
This competition aims to foster creativity and technical expertise in the realm of SNNs. We encourage you to push the boundaries of innovation and deliver groundbreaking solutions.
Best of luck!
Sincerely,
The team at SECQAI
ai.competition@secqai.com