A Framework for Unsupervised Deep Clustering Contribution Request #32
Replies: 12 comments 13 replies
-
|
Hello, Before I jump into that, I thought it would be beneficial to start with less complicated data. Maybe that is an ignorant thing to say as I am not super familiar with EEG data. Either way, I thought this would be a great opportunity to 1. contribute to an open source community (I have always been passionate about "free code") and 2. do something that will help me in my own research. I am new to autoencoders, so I wrote this notebook that trains an autoencoder on MNIST data to better understand the idea. I was actually surprised how simple autoencoders are. I Also cluster the embedded images and visualize them. I chose 10 clusters for obvious reasons, but if you look at the figure in the notebook you see that one of the clusters group the bold numbers not necessarily a 0-9 digit. I was wondering If I could try it with EEG data. Does this org have any that I can use? If not, I can just download the SEED or DEAP datasets. Next on the list for me is going to try and add a clustering step inside the autoencoder and create some custom loss function that includes the clustering loss. I also might use Pytorch rather than Tensorflow/keras or make my own from scratch using JAX. If anyone has any other ideas to try, let me know. I need to spend some more time in the literature. cbraxtonowens (at) gmail.com |
Beta Was this translation helpful? Give feedback.
-
|
With autoencoders you are calculating a loss using the decoded output with the input. Basically, your encoder maps the input to a latent space, then the decoder tries to map it back to look like the input. In a simple autoencoder you just want to minimize the error between your decoded output and the original input. You can add a regularization term that also minimizes the distance from a precalculated cluster center. Essentially you are just adding a prior that assumes it is important to be close to a cluster center. |
Beta Was this translation helpful? Give feedback.
-
|
I get what you are suggesting, but i think we are starting from different assamptions, the need for a clustering comes from the fact that the data are unlabeled and also there is no clear way to do it (correct me if i am wrong), even the clusters resulting will not necessary have a semantic meaning(eg. similar brain areas activated simultaneusly). At a certain point of your process there will be the need for labeling the data, assign them to a certain cluster, but i am not sure how you can do it with a meaning and if you can do it, why not do it from the start and transmute to a classification task the whole thing. |
Beta Was this translation helpful? Give feedback.
-
|
Absolutely, I understand your concerns and the points you've raised. I think the org is better prepared to address your concern. One reason to employ clustering in this context is to explore the dataset and uncover any inherent relationships or patterns that exist within the unlabeled data. While it's true that these clusters might not immediately align with predefined, semantically meaningful categories, the relationships they reveal can be extremely valuable. These discovered relationships can offer new insights or hypotheses for researchers to investigate further. Clustering allows us to identify these natural groupings in the data without the bias of predefined labels. This exploratory step can uncover previously unknown connections or similarities between data points, which could be crucial for understanding complex datasets, such as those involving brain activity. Once we've identified these relationships, the insights gained can guide the creation of more meaningful labels or even inform the development of a classification model. |
Beta Was this translation helpful? Give feedback.
-
|
Good discussion here. I also draw your attention to the fact that in the medical and healthcare domain, labels are scarce, expensive, and noisy (inter-rater variability). |
Beta Was this translation helpful? Give feedback.
-
|
Thanks for the reply @braxtonowens. |
Beta Was this translation helpful? Give feedback.
-
|
I will ask here some clarification, tell me if this is not the right place. |
Beta Was this translation helpful? Give feedback.
-
|
most of the clinical studies with a very low/limited number of subjects/data samples. |
Beta Was this translation helpful? Give feedback.
-
|
in order to solve this the idea is to use transfer learning/fine-tuning from open source models, without accounting for any limitation on the computational power side, am i right? |
Beta Was this translation helpful? Give feedback.
-
|
At this stage yes. But after the proof of concept, we will aim for implementation on edge devices. |
Beta Was this translation helpful? Give feedback.
-
|
Have you included your email? |
Beta Was this translation helpful? Give feedback.
-
|
my email is manuel.db02@gmail.com |
Beta Was this translation helpful? Give feedback.
-
|
Greetings everyone, My name is J Damour Nsanzimfura, and I am currently pursuing a Master's of Science in Engineering with a focus on Artificial Intelligence at Carnegie Mellon University in Kigali. With a strong foundation in deep learning, data structures, and algorithms, I have gained practical experience through internships and personal projects that have honed my skills in software engineering and machine learning. You can reach out to me at ndamour@andrew.cmu.edu/ |
Beta Was this translation helpful? Give feedback.
-
|
Hello everyone, My name is Avdhesh Varshney. I am a GSOC'24 Contributor.
I would like to work on the Project Idea 4 - A Framework for Unsupervised Deep Clustering. Connect with me: |
Beta Was this translation helpful? Give feedback.
-
|
Hello everyone, My name is Shashank Shekhar Singh, and I'm currently a second-year student at the Indian Institute of Technology (BHU), Varanasi. I am actively involved in various open-source contributions, particularly in the field of Machine Learning and Deep Learning. Some of my notable contributions can be seen using my GitHub ID. As a passionate contributor, I am excited about the opportunity to participate in Google Summer of Code 2024 as a contributor. I am well-versed in ML and DL tech stacks, and I am keen to collaborate on Project Idea 4 - A Framework for Unsupervised Deep Clustering. If the mentor wants to share some insights for this project idea or anyone else wants to collaborate on it, I am more than willing to work together and support each other's growth. Looking forward to connecting with like-minded individuals and making meaningful contributions to the project. Best regards, |
Beta Was this translation helpful? Give feedback.
-
|
Hello folks, my name is Claude Kwizera, I am a student at Carnegie Mellon University in a field of AI. My background is in Software engineering with experience of 5 years developing backend, frontend and mobile applications. I am currently pursuing my degree in AI as it has become my passion so since 3 years ago. My interests are in Computer vision, NLP and Neural Science. I am considering working on an open project to challenge myself with this project so that I can bring out my zeal to contribute something and gain more experience as well. my contacts are the following if you would like to connect: |
Beta Was this translation helpful? Give feedback.
-
|
Hello, I have a background in working with EEG data, which I gained through my participation in the PhysioNet Challenge - 2023. You can find the paper about that here. During the challenge, I worked with the preprocessing of EEG data, and apart from the methods mentioned in that paper, I tried alternative approaches like wavelet transformations, and 1D convolution with pyramid pooling to handle varying lengths of EEG data for each patient. Mail: chay5522kalyan (at) gmail.com |
Beta Was this translation helpful? Give feedback.
-
|
Hello everyone, Nice to meet you all! I’m Yuru Jing (Bertie) from China, and I recently completed my postgraduate studies in Data Science and Machine Learning at University College London. Prior to this, I obtained a Bachelor's degree in Mathematics with a minor in Finance from institutions in Wisconsin and Illinois, graduating in 2020. During my last two years of undergraduate, I delved into machine learning and mathematical modelling, particularly focusing on its applications in biomedicine and healthcare. I spearheaded projects such as machine learning clustering for Iris Species in Intercollegiate Biomathematics Alliance and conducted research and dissertation on Statistical Modeling of SARS-Cov-2 Mutation which I presented at the 2021 International Symposium on Biomathematics and Ecology Education and Research. In my postgraduate studies, I refined my expertise in AI for healthcare, contributing to publications such as "Machine Learning Performance Analysis to Predict Stroke" and "Auto-outlier Fusion Technique for Chest X-ray Classification." Moreover, I collaborated with Moorfields Eye Hospital NHS Foundation Trust for my final dissertation, delving into "New Designs to Predict Refractive Error From Retinal Fundus Images using Deep Learning." I found great passion in addressing complex healthcare challenges, including multimodal features extraction and fusion, semi-supervised clustering, and imbalanced classification. I am particularly drawn to the "A Framework for Unsupervised Deep Clustering" project, and I believe my related skills and experiences as follows would be highly beneficial to make a contribution fot this project. Highlights: Deconvolution-based image restoration and Neural Spike Simulation Github: https://github.com/YuruJing |
Beta Was this translation helpful? Give feedback.
-
|
My name is Jivisha Gour, and I am an undergraduate currently pursuing my B.Tech in Computer Science at JK Lakshmipat University, Jaipur. I am thrilled to propose my involvement in advancing the unsupervised deep clustering framework for EEG data analysis through this program, under the guidance of mentors Mahmoud Zeydabadinezhad and Babak Mahmoudi, PhD. My passion for leveraging cutting-edge techniques to unlock insights from a dataset drives my eagerness to contribute to this project. With a solid foundation in deep learning, machine learning, and artificial intelligence, bolstered by completion of three specialized courses in deep learning, one in machine learning, and one in AI, I have honed my skills in tackling complex data analysis tasks. My practical experience extends to real-world projects, including "Life Expectancy," (https://github.com/JivishaGour/LifeExpectancyProject) where I delved into the intricacies of global health disparities by employing various machine learning techniques to analyze life expectancy trends across different countries. Additionally, my project "Life Underwater" showcased my proficiency in Python and statistical analysis, as I explored the Sustainable Development Goals (SDGs) related to aquatic ecosystems using Python libraries and statistical methods. These projects not only underscore my expertise in leveraging machine learning for data-driven insights but also demonstrate my ability to apply these techniques to diverse domains, setting a strong foundation for contributing to the development of advanced unsupervised deep clustering techniques for EEG data analysis. **Specifically, I aim to focus on algorithm development, model optimization, and documentation, ensuring the framework's effectiveness and accessibility. ** I'm excited to propose some innovative methods that could elevate our unsupervised deep clustering framework:
Leveraging my proficiency in Python, TensorFlow, and data preprocessing techniques, I am well-equipped to make meaningful contributions to the project. I am committed to engaging with domain experts to validate the framework's efficacy in real-world healthcare applications, such as epilepsy diagnosis and brain-computer interface development. I look forward to collaborating with the mentors and the community to make a meaningful impact in this vital area of research. Email: jivishagaur@gmail.com |
Beta Was this translation helpful? Give feedback.
-
|
Hello everyone, I'm Utkarsh Singh Gehlot, a third-year undergraduate student pursuing my B.Tech in Computer Science at JK Lakshmipat University. I'm excited to contribute to the development of an unsupervised deep clustering framework for EEG data analysis. My journey in technology has been a blend of hands-on experience and continuous learning. Currently, I'm expanding my knowledge in advanced machine learning techniques, including deep learning clustering techniques, through self-study and coursework. This ongoing learning journey has provided me with insights into state-of-the-art clustering algorithms such as autoencoder-based clustering, spectral clustering, and deep embedded clustering. During my internship at IIIT Bangalore, I had the privilege to immerse myself in dynamic data clustering techniques using Topological Data Analysis (TDA). This experience not only honed my skills in algorithm development but also ignited my curiosity to delve deeper into the complexities of data analysis. My research focused on developing innovative algorithms for dynamic clustering and analyzing cluster lifecycles, showcasing my ability to tackle complex algorithmic challenges and contribute to cutting-edge research. The project's objective to develop an open-source framework utilizing unsupervised deep-learning techniques for data clustering aligns perfectly with my interests and aspirations. Leveraging my experience at and my ongoing learning in deep learning clustering techniques, I am eager to contribute to the development of robust clustering algorithms for EEG data analysis. By combining my practical experience with my theoretical knowledge, I am confident in my ability to contribute meaningfully to the project. I am committed to exploring innovative techniques, such as active learning and transfer learning, to enhance the framework's performance and scalability. Looking ahead, I see immense potential in the outcomes of this project to drive advancements in medical data analysis. By deploying an open-source implementation of the framework and exploring its applications in various healthcare domains, we have the opportunity to make a tangible impact on patient care and treatment outcomes. I am excited to collaborate with mentors, fellow contributors, and domain experts to bring this project to fruition and contribute to the advancement of data-driven solutions in healthcare. Email: utkarshsinghgehlot952@gmail.com |
Beta Was this translation helpful? Give feedback.
-
|
Hi Folks!! A brief background of me: I am excited to collaborate with mentors, fellow contributors, and domain experts to make an impact in the area of Healthcare. |
Beta Was this translation helpful? Give feedback.
-
|
hello guys! |
Beta Was this translation helpful? Give feedback.
-
Hi Everyone,
My name is Mohamed I study systems and biomedical engineering, and I have great passion for contributing to this project,I am a machine learning contributor at Omdena.I am looking forward to get myself challenged working on this project. Thanks
Beta Was this translation helpful? Give feedback.
All reactions