Infolab@SKKU
LightNews LightNews
banner
infolab.bsky.social
Infolab@SKKU
@infolab.bsky.social
23 followers 5 following 9 posts
InfoLab is a research group at SKKU pushing the boundaries of security and machine learning, especially in bioinformatics and biomedical discovery. For more information, please visit our website at infolab.skku.edu
Posts Replies Media Videos
infolab.bsky.social
Top AI pioneers now say human-level general intelligence isn’t a future vision, it’s happening now. Are we ready for an era where machines truly rival humans?

#general_Intelligence #AGI #FT #AI_submit

youtu.be/0zXSrsKlm5A?...
The Minds of Modern AI: Jensen Huang, Geoffrey Hinton, Yann LeCun & the AI Vision of the Future
YouTube video by FT Live
youtu.be
November 8, 2025 at 2:02 AM
infolab.bsky.social
Study shows that humans and neural networks face the same trade-off when learning multiple tasks: learning more similar tasks boosts “transfer” but increases “interference”.
Understanding this helps optimise continual learning in both brains and machines.

Read more www.nature.com/articles/s41...
Humans and neural networks show similar patterns of transfer and interference during continual learning - Nature Human Behaviour
When learning new tasks, both humans and artificial neural networks face a trade-off between reusing prior knowledge to learn faster and avoiding the disruption of earlier learning. This study shows t...
www.nature.com
November 2, 2025 at 4:53 AM
infolab.bsky.social
How will Large Language Models reshape science?
Nature Computational Science’s Focus issue dives into breakthroughs, challenges & the future of AI in research.

Read here www.nature.com/collections/...

#AI #LLM #Science #Research #FutureOfAI #Nature
The impact of large language models in science
Large language models (LLMs) are rapidly being implemented in a wide range of disciplines, with the promise of unlocking new possibilities for scientific ...
www.nature.com
September 28, 2025 at 8:20 AM
infolab.bsky.social
A framework to build AI scaling laws for cost-efficient LLM training, helping teams get the most out of limited compute budgets.

#LLM #AITech #AIEfficiency #ScalingLaws #ArtificialIntelligence

💡 Key insight: smarter scaling = more performance per dollar.
Read more 👇
news.mit.edu/2025/how-bui...
How to build AI scaling laws for efficient LLM training and budget maximization
MIT and MIT-IBM Watson AI Lab researchers have developed a universal guide for estimating how large language models (LLMs) will perform based on smaller models in the same family.
news.mit.edu
September 24, 2025 at 12:20 PM
Reposted by Infolab@SKKU
dppapers.bsky.social
Differentially Private Federated Clustering with Random Rebalancing

Xiyuan Yang, Shengyuan Hu, Soyeon Kim, Tian Li

http://arxiv.org/abs/2508.06183
Differentially Private Federated Clustering with Random Rebalancing Xiyuan Yang, Shengyuan Hu, Soyeon Kim, Tian Li http://arxiv.org/abs/2508.06183 Federated clustering aims to group similar clients into clusters and produce one model for each cluster. Such a personalization approach typically improves model performance compared with training a single model to serve all clients, but can be more vulnerable to privacy leakage. Directly applying client-level differentially private (DP) mechanisms to federated clustering could degrade the utilities significantly. We identify that such deficiencies are mainly due to the difficulties of averaging privacy noise within each cluster (following standard privacy mechanisms), as the number of clients assigned to the same clusters is uncontrolled. To this end, we propose a simple and effective technique, named RR-Cluster, that can be viewed as a light-weight add-on to many federated clustering algorithms. RR-Cluster achieves reduced privacy noise via randomly rebalancing cluster assignments, guaranteeing a minimum number of clients assigned to each cluster. We analyze the tradeoffs between decreased privacy noise variance and potentially increased bias from incorrect assignments and provide convergence bounds for RR-Clsuter. Empirically, we demonstrate the RR-Cluster plugged into strong federated clustering algorithms results in significantly improved privacy/utility tradeoffs across both synthetic and real-world datasets.
August 11, 2025 at 3:48 AM
infolab.bsky.social
Apple releases Embedding Atlas — an open-source tool to explore & compare large-scale embeddings with ease. Ideal for research: visualize, cluster, and search embeddings in your browser via WebGPU. #MachineLearning #AIResearch

GitHub
github.com/apple/embedd...
GitHub · Build and ship software on a single, collaborative platform
Join the world's most widely adopted, AI-powered developer platform where millions of developers, businesses, and the largest open source community build software that advances humanity.
github.com
August 10, 2025 at 2:40 AM
infolab.bsky.social
MLE-STAR agent system by Google to automate complex machine learning ML pipeline, a true leap in the automation of machine learning engineering

github.com/google/adk-s...

#GoogleAI #MLEGEN #AItools #AutoML #AIAgent
adk-samples/python/agents/machine-learning-engineering at main · google/adk-samples
A collection of sample agents built with Agent Development (ADK) - google/adk-samples
github.com
August 4, 2025 at 2:42 PM
infolab.bsky.social
New large-scale study finds: LLMs fail to re-identify the same bug 78% of the time after harmless code edits. Most models don’t really understand code, they just pattern-match.
arxiv.org/abs/2504.04372

#AI #LLM #CodeIntelligence #AIResearch #Debugging #SoftwareEngineering
How Accurately Do Large Language Models Understand Code?
Large Language Models (LLMs) are increasingly used in post-development tasks such as code repair and testing. A key factor in these tasks' success is the model's deep understanding of code. However, t...
arxiv.org
July 9, 2025 at 7:16 PM
infolab.bsky.social
New AI research warns: LLM reasoning traces may leak private user data!

Privacy in AI isn't just about output — it's in the process too.

📖 www.marktechpost.com/2025/06/25/n...

#AI #LLM #Privacy #DataSecurity #ChainOfThought #AIResearch #ResponsibleAI
New AI Research Reveals Privacy Risks in LLM Reasoning Traces
New study exposes how reasoning traces in large language models leak sensitive user data, posing significant contextual privacy risks
www.marktechpost.com
July 7, 2025 at 4:29 PM
infolab.bsky.social
ETH Zurich & Stanford researchers just dropped MIRIADE — a massive 5.8M-pair dataset to boost LLM reasoning accuracy in medical AI!

📚 www.marktechpost.com/2025/06/25/e...

#AI #MedicalAI #LLM #HealthcareInnovation #MIRIADE #StanfordAI #ETHZurich
ETH and Stanford Researchers Introduce MIRIAD: A 5.8M Pair Dataset to Improve LLM Accuracy in Medical AI
ETH and Stanford launch MIRIAD, a 5.8M medical QA dataset designed to boost LLM accuracy and reduce hallucinations in healthcare
www.marktechpost.com
July 7, 2025 at 4:28 PM
Reposted by Infolab@SKKU
simonsinstitute.bsky.social
In a recent talk at the Simons Institute, Somesh Jha made a case for applying a security and cryptography mindset to evaluating the trustworthiness of machine learning systems, particularly in adversarial and privacy-sensitive contexts.

bit.ly/43Xe0AP
Safety of GenAI through the Lens of Security and Cryptography
In this deliberately provocative two-part talk from the recent workshop on Theoretical Aspects of Trustworthy AI, Somesh Jha (University of Wisconsin) makes a case for applying a security and cryptogr...
bit.ly
May 28, 2025 at 7:25 AM
Reposted by Infolab@SKKU
ai-firehose.column.social
New research reveals a vulnerability in language models, enabling knowledge manipulation through subtle feedback, injecting false information and vulnerabilities into code generation, posing risks for AI applications reliant on user-driven preference tuning. https://arxiv.org/abs/2507.02850
LLM Hypnosis: Exploiting User Feedback for Unauthorized Knowledge Injection to All Users
ArXiv link for LLM Hypnosis: Exploiting User Feedback for Unauthorized Knowledge Injection to All Users
arxiv.org
July 4, 2025 at 5:40 PM
Reposted by Infolab@SKKU
ai-firehose.column.social
Researchers have developed ExpProof, a new protocol using Zero-Knowledge Proofs to create trustworthy explanations for confidential machine learning models, ensuring model integrity and correct explanations in adversarial settings without losing confidentiality. https://arxiv.org/abs/2502.03773
ExpProof : Operationalizing Explanations for Confidential Models with ZKPs
ArXiv link for ExpProof : Operationalizing Explanations for Confidential Models with ZKPs
arxiv.org
June 2, 2025 at 11:20 PM
Reposted by Infolab@SKKU
rflamary.bsky.social
The most important aspect when facing data shift is the type of shift present in the data. I will give below a few examples of shifts and some existing methods to compensate for it.🧵1/6
July 1, 2025 at 9:39 AM
Reposted by Infolab@SKKU
compscioxford.bsky.social
Did you know computer science plays a vital role in many kinds of research? From methane-detecting satellites & digital models of human hearts, to using AI to improve pandemic preparedness or track leopards & listen to whales. #ResearchAppreciationDay
A dark blue graphic with bright blue text reading ‘Research Appreciation Day’. Underneath this there is a bright blue graphic of lines and dots that looks like a computer board. Coming off the graphic, there is text reading 'Algorithms and Complexity, Artificial Intelligence and Machine Learning, Automated Verification, Computational Biology and Health Informatics, Data, Knowledge and Action, Human Centred Computing, Programming Languages, Quantum, Security, Systems'.
July 4, 2025 at 9:36 AM
Reposted by Infolab@SKKU
dppapers.bsky.social
Federated Learning: A Survey on Privacy-Preserving Collaborative Intelligence

Edward Collins, Michel Wang

http://arxiv.org/abs/2504.17703
Federated Learning: A Survey on Privacy-Preserving Collaborative Intelligence Edward Collins, Michel Wang http://arxiv.org/abs/2504.17703 Federated Learning (FL) has emerged as a transformative paradigm in the field of distributed machine learning, enabling multiple clients such as mobile devices, edge nodes, or organizations to collaboratively train a shared global model without the need to centralize sensitive data. This decentralized approach addresses growing concerns around data privacy, security, and regulatory compliance, making it particularly attractive in domains such as healthcare, finance, and smart IoT systems. This survey provides a concise yet comprehensive overview of Federated Learning, beginning with its core architecture and communication protocol. We discuss the standard FL lifecycle, including local training, model aggregation, and global updates. A particular emphasis is placed on key technical challenges such as handling non-IID (non-independent and identically distributed) data, mitigating system and hardware heterogeneity, reducing communication overhead, and ensuring privacy through mechanisms like differential privacy and secure aggregation. Furthermore, we examine emerging trends in FL research, including personalized FL, cross-device versus cross-silo settings, and integration with other paradigms such as reinforcement learning and quantum computing. We also highlight real-world applications and summarize benchmark datasets and evaluation metrics commonly used in FL research. Finally, we outline open research problems and future directions to guide the development of scalable, efficient, and trustworthy FL systems.
April 25, 2025 at 3:48 AM
Reposted by Infolab@SKKU
synthemaeu.bsky.social
🚀 #SYNTHEMA + flower.ai = a new era for secure federated learning in healthcare!

With privacy & security at the core, we’re leveraging Flower’s open-source framework to enhance #AI model training—without sharing raw data.

👉 Learn more: synthema.eu/2025/02/17/s...

#HealthcareTech
February 18, 2025 at 8:55 AM
Reposted by Infolab@SKKU
arxiv-cs-cr.bsky.social
Rob Romijnders, Stefanos Laskaridis, Ali Shahin Shamsabadi, Hamed Haddadi
NoEsis: Differentially Private Knowledge Transfer in Modular LLM Adaptation
https://arxiv.org/abs/2504.18147
April 28, 2025 at 4:48 AM
Reposted by Infolab@SKKU
arxiv-cs-cr.bsky.social
Abrar Fahim, Shamik Dey, Md. Nurul Absur, Md Kamrul Siam, Md. Tahmidul Huque, Jafreen Jafor Godhuli
Optimized Approaches to Malware Detection: A Study of Machine Learning and Deep Learning Techniques
https://arxiv.org/abs/2504.17930
April 28, 2025 at 5:52 AM
Reposted by Infolab@SKKU
arxiv-cs-cr.bsky.social
Firuz Juraev, Mohammed Abuhamad, Eric Chan-Tin, George K. Thiruvathukal, Tamer Abuhmed
From Attack to Defense: Insights into Deep Learning Security Measures in Black-Box Settings
https://arxiv.org/abs/2405.01963
May 6, 2024 at 5:13 PM
Reposted by Infolab@SKKU
arxiv-cs-cv.bsky.social
Firuz Juraev, Mohammed Abuhamad, Simon S. Woo, George K Thiruvathukal, Tamer Abuhmed
Impact of Architectural Modifications on Deep Learning Adversarial Robustness
https://arxiv.org/abs/2405.01934
May 6, 2024 at 6:15 PM
Reposted by Infolab@SKKU
dppapers.bsky.social
Local Differential Privacy is Not Enough: A Sample Reconstruction Attack against Federated Learning with Local Differential Privacy
Zhichao You, Xuewen Dong, Shujun Li, Ximeng Liu, Siqi Ma, Yulong Shen
http://arxiv.org/abs/2502.08151
Local Differential Privacy is Not Enough: A Sample Reconstruction Attack against Federated Learning with Local Differential Privacy Zhichao You, Xuewen Dong, Shujun Li, Ximeng Liu, Siqi Ma, Yulong Shen http://arxiv.org/abs/2502.08151 Reconstruction attacks against federated learning (FL) aim to reconstruct users' samples through users' uploaded gradients. Local differential privacy (LDP) is regarded as an effective defense against various attacks, including sample reconstruction in FL, where gradients are clipped and perturbed. Existing attacks are ineffective in FL with LDP since clipped and perturbed gradients obliterate most sample information for reconstruction. Besides, existing attacks embed additional sample information into gradients to improve the attack effect and cause gradient expansion, leading to a more severe gradient clipping in FL with LDP. In this paper, we propose a sample reconstruction attack against LDP-based FL with any target models to reconstruct victims' sensitive samples to illustrate that FL with LDP is not flawless. Considering gradient expansion in reconstruction attacks and noise in LDP, the core of the proposed attack is gradient compression and reconstructed sample denoising. For gradient compression, an inference structure based on sample characteristics is presented to reduce redundant gradients against LDP. For reconstructed sample denoising, we artificially introduce zero gradients to observe noise distribution and scale confidence interval to filter the noise. Theoretical proof guarantees the effectiveness of the proposed attack. Evaluations show that the proposed attack is the only attack that reconstructs victims' training samples in LDP-based FL and has little impact on the target model's accuracy. We conclude that LDP-based FL needs further improvements to defend against sample reconstruction attacks effectively.
February 13, 2025 at 4:33 AM
Reposted by Infolab@SKKU
ramikrispin.bsky.social
(1/3) Apple open source a new Python library for simulation framework for accelerating research in Private Federated Learning. 🧵👇🏼

#datascience #machinelearning #python #deeplearning
March 4, 2024 at 2:16 PM
Reposted by Infolab@SKKU
azoai.bsky.social
Researchers Address Key Challenges in Federated Learning 🔍🤖📊 www.azoai.com/news/2024120... #FederatedLearning #MachineLearning #DataPrivacy #AIResearch #ModelPersonalization #FLChallenges #TechInnovation #DataScience #AITech #SecureData
Researchers Address Key Challenges in Federated Learning
Researchers analyze federated learning challenges like data diversity and resource constraints, offering experimental insights to enhance privacy, efficiency, and scalability.
www.azoai.com
December 9, 2024 at 2:54 AM
Reposted by Infolab@SKKU
tomsilico.bsky.social
Also note this alternative federated learning approach talk at the RDKit UGM: www.youtube.com/watch?v=Y1mj...
paper: pubs.acs.org/doi/10.1021/...
Andrea Andrews-Morger: Federated learning in computational toxicology
YouTube video by RDKit
www.youtube.com
November 19, 2024 at 1:51 AM