#30daysofflcode search results

I am publicly taking the #30daysofFLCode challenge with @openmind_agi . I hope I can share all my learnings day wise here. #FederatedLearning #FL

I'm publicly committing to the #30DaysOfFLCode Challenge! Join me in learning more about #FederatedLearning → 30DaysOfFLCode.com

I'm publicly committing to the #30DaysOfFLCode Challenge! Join me in learning more about #FederatedLearning → 30DaysOfFLCode.com

SyftBox : Compute on private data without accessing it! 🔹 Data Owners keep control & share only results 🔹 API Developers run secure computations 🔹 Fully decentralized & privacy-first syftbox-documentation.openmined.org/tutorials/comp… #30DaysOfFLCode #AI #Privacy #Decentralization #SyftBox

I'm publicly committing to the #30DaysOfFLCode Challenge! Join me in learning more about #FederatedLearning and Privacy Enhancing Technologies (PETs). Let’s build our skills and shape the future of AI together! 🚀 Check out here: 30DaysOfFLCode.com #AI #OpenMined

I'm publicly committing to the #30DaysOfFLCode Challenge! Join me in learning more about #FederatedLearning → 30DaysOfFLCode.com

Last month, I completed the #30DaysOfFLCode Challenge by @openminedorg! I encourage anyone interested in learning about privacy-preserving technologies, security in AI, and FL to take the challenge! info.openmined.org/30daysofflcode

Day 30: Efficient Pruning for Machine Learning under HE -Pruning boosts traditional ML efficiency but not in HE-based ML. Tile sparsity enables efficient efficient pruning by skipping entire blocks of encrypted data, reducing operations. fhe.org/meetups/041-Ef… #30DaysOfFLCode

Day 29: Concrete ML - Machine Learning on Encrypted Data A model is converted to an FHE-compatible format via quantization, and its weights are encrypted. It then performs predictions on FHE-encrypted data, with the results being encrypted youtube.com/watch?v=DP_4OB… #30DaysOfFLCode

Day 28: Verifiable FHE 1 Users can employ ZKP to verify that the FHE encryption of their plain query was correctly done. On the server side, it can be used to verify that the computations on this encrypted data are carried out correctly. blog.sunscreen.tech/snarks-shortco… #30DaysOfFLCode

Day 27: Apple's Enhanced Visual Search with ML, HE & PNNS ML model on your device identifies matching landmarks in your photo gallery, creates an encrypted code from them which is used with HE to retrieve a desired photo from Apple's server machinelearning.apple.com/research/homom… #30DaysOfFLCode

Day 26: Private Information Retrieval PIR allows a user to send an encrypted query to a server, which then performs homomorphic computations to retrieve the desired information, sending back the result still encrypted to preserve privacy. machinelearning.apple.com/research/homom… #30DaysOfFLCode

Day 25: Problems and Research Trends in Federated Learning I read about problems in FL which requires further research. This includes improving efficiency and effectiveness, reduction in training biases, and robustness against attacks. drive.google.com/file/d/1QGY2Zy… #30DaysOfFLCode

Day 24: Cross-Silo Federated Learning -It involves fewer clients (organizations) with specific identities -Most clients participate in each FL round with very few going offline. -Data Partition can be vertical or horizontal across clients. drive.google.com/file/d/1QGY2Zy… #30DaysOfFLCode

Day 23: Cross-device Federated Learning (FL) and Challenges - In cross-device FL, devices share the same features but have different data samples - Communication among FL clients is challenging because the mobile devices often go offline. drive.google.com/file/d/1QGY2Zy… #30DaysOfFLCode

Day 22: Cross-Device Federated Learning at Apple Apple uses differentially private federated learning to personalize Siri. This enables an iPhone to respond exclusively to its owner’s voice, even in the presence of other phones and voices technologyreview.com/2019/12/11/131… #30DaysOfFLCode

Day 21: Privacy Preserving AI - Lecture (Andrew Trask - OpenMined) Focuses on how privacy enhancing technologies such as differential privacy, multiparty computation and federated learning enables remote training of models on private data.youtu.be/4zrU54VIK6k?fe… #30DaysOfFLCode

Completed the #30DaysOfFLCode Learned so much and advanced mixing of PHEs project. Next up: zk proofs for verifiable computations to enhance PETs. Grateful for the journey and the amazing people I met! 🚀 the journey will continue …

Day 20: Beyond Privacy Trade-offs with Structured Transparency Participants can't prevent other collaborators from misusing their data (copy problem). The proposed techniques enable collaborators to reduce the risks of collaboration. arxiv.org/abs/2012.08347 #30DaysOfFLCode

Day 19: Federated fine-tuning of LLMs (Part 3) I continued with the decentralized/federated fine-tuning of the LLM, leveraging parameter efficient fine-tuning to optimize communication efficiency and differential privacy to improve privacy learn.deeplearning.ai/courses/intro-… #30DaysOfFLCode

#30DaysOfFLCode

day 16 of #30DaysOfFLCode: continued working through the Syft tutorial. trained a simple ml model on mock cancer data, created & submitted a research proposal and code request to the datasite and tested that remote code execution wouldn't work before research project approval.

I will be joining #30daysOfFLCode @openminedorg

day 4 of #30DaysOfFLCode: reviewed syftbox's computational model. then installed a fun project: a cpu tracker API. it gathers cpu data from participating peers, aggregates it & visualizes it. also reviewed a foundational paper on FL by Li et al here: arxiv.org/pdf/1908.07873

#30DaysOfFLCode

day 18 of #30DaysOfFLCode: completed the fl cancer tutorial today. ran the model locally as a data admin, then approved the code request to the data. accessed the approved project as the data scientist & ran the model on the actual data. pretty fun, now onto more SyftBox :)

day 31 of #30DaysOfFLCode: challenge complete :) thanks @openminedorg 🌟 enjoyed a show & tell session with the team to celebrate this milestone. presentations included a FedRAG, ring computations, the syftbox netflix app and more. this is just the beginning. more fl to come!

Day 30 of #30DaysOfFLCode! Final day of the challenge! I explored a real use case called FedJudge. It alleviates the red flag of central training in the Legal Intelligence field. On top of that, it uses parameter efficient fine-tuning to further increase Legal LLMs efficiency.

Day 29 of #30DaysOfFLCode! I reviewed "Analysis of Privacy Leakage in Federated LLMs." FL research has made adjustments to adapt with LLMs, but privacy analysis is missing. The paper proposes two active MI attacks and provides guaranteed theoretical success rates on four LLMs.

Day 28 of #30DaysOfFLCode! I covered FedSecurity, a benchmark of attacks and defenses crucial to FL research. It's a customizable API that includes attacks/defenses before, during, and after aggregation, which is not included in previous benchmarks. It also extends to FedLLMs!

Day 26 of #30DaysOfFLCode! I discovered a new reconstruction attack, LOKI, intended to be an improvement on attacks presented in Robbing the Fed and When the Curious Abandon Honesty. Following my own work, OASIS, I was able to mitigate the attack with simple data augmentation.

Day 22 of #30DaysOfFLCode! Continuing with Flower, I ran into FlowerTune LLM where the 7B parameter LLaMA was quantized to 3B parameters and was still able to perform on par! I'm more interested in how quantizing a federated model affects fairness in real-world scenarios.

Day 29 of #30DaysOfFLCode Exploring FL workflow with SyftBox Read more: syftbox-documentation.openmined.org/tutorials/fede… #DifferentialPrivacy @openminedorg

Day 15: #30DaysOfFLCode Had a great time with Openmined First "Show & Tell" Session, where we got the opportunity to learn from other members FL projects. First session is an introductory FL computation on SyftBox by Ionesio Junior, then an exciting e-voting project by Lucas…

Day 24 of #30DaysOfFLCode with @openminedorg! I reviewed the paper Mixed-Precision Quantization for FL on Resource-Constrained Heterogeneous Devices. The authors claim to develop FedMPQ, where the mixed-precision quantization scheme occurs at different layers in an FL setup.

I'm committing to the #30DaysOfFLCode Challenge! Join me in learning more about #FederatedLearning → 30DaysOfFLCode.com

Day-19 of #30DaysOfFLCode Worked on the federated rag project and had a fruitful discussion with the whole team about progress so far and plan of action! excited to see it all take good shape!!

Day 1 of #30DaysOfFLCode ! I'm taking on a challenge to dive deeper into my research area with @openminedorg. I'm excited to cover new literature around privacy in FL and all the different attack vectors. I started by reviewing papers that cover active gradient inversion!

I'm publicly committing to the #30DaysOfFLCode Challenge! Join me in learning more about #FederatedLearning → 30DaysOfFLCode.com @openminedorg #PrivateAI #FedratedLearning #DifferentialPrivacy

Day 1: Completed SyftBox tutorial & intro to Federated Learning (FL)! 🚀 Explored its modular framework for privacy-preserving tech. Deployed a CPU Tracker with Differential Privacy. Learned FL concepts: training models collaboratively without sharing data. #30DaysOfFLCode