A naukri.com initiative
ML News
Arxiv
3d
256
Image Credit: Arxiv
Symmetry-Informed Graph Neural Networks for Carbon Dioxide Isotherm and Adsorption Prediction in Aluminum-Substituted Zeolites
- Accurately predicting adsorption properties in nanoporous materials using Deep Learning models remains a challenging task.
- SymGNN is a graph neural network architecture that leverages material symmetries to improve adsorption property prediction.
- The model successfully captures key adsorption trends, including the influence of both the framework and aluminium distribution on CO2 adsorption.
- The study suggests promising directions for fine-tuning with experimental data and generative approaches for the inverse design of multifunctional nanomaterials.
Read Full Article
15 Likes
Arxiv
3d
200
Image Credit: Arxiv
Malicious and Unintentional Disclosure Risks in Large Language Models for Code Generation
- This paper explores the risks of unintentional and malicious disclosure in large language models trained for code generation.
- Unintentional disclosure refers to the language model presenting secrets to users without user intent, while malicious disclosure refers to presenting secrets to an attacker.
- The study assesses the risks of unintentional and malicious disclosure in the Open Language Model (OLMo) family of models and the Dolma training datasets.
- The results show that changes in data source and processing greatly affect the risk of unintended memorization, and the risk of disclosing sensitive information varies based on prompt strategies and types of sensitive information.
Read Full Article
12 Likes
Arxiv
3d
216
Image Credit: Arxiv
Boosting Large Language Models with Mask Fine-Tuning
- A new language model fine-tuning paradigm called Mask Fine-Tuning (MFT) has been introduced.
- MFT breaks the integrity of the model to improve its performance.
- Extensive experiments show consistent performance boosts across various domains and backbones.
- MFT extends the functionality of mask learning for model compression to a more general scope.
Read Full Article
12 Likes
Arxiv
3d
336
Image Credit: Arxiv
Congenital Heart Disease Classification Using Phonocardiograms: A Scalable Screening Tool for Diverse Environments
- Congenital heart disease (CHD) is a critical condition that demands early detection, particularly in infancy and childhood.
- A deep learning model designed to detect CHD using phonocardiogram (PCG) signals achieved high accuracy of 94.1%, sensitivity of 92.7%, specificity of 96.3%.
- The model demonstrated robust performance on diverse datasets from Bangladesh, as well as public datasets, showing its generalizability to different populations and data sources.
- The research suggests that an AI-driven digital stethoscope could be a cost-effective screening tool for CHD in resource-limited settings, improving patient outcomes.
Read Full Article
20 Likes
Arxiv
3d
360
Image Credit: Arxiv
Policy Optimization and Multi-agent Reinforcement Learning for Mean-variance Team Stochastic Games
- Researchers study a long-run mean-variance team stochastic game (MV-TSG) and propose a Mean-Variance Multi-Agent Policy Iteration (MV-MAPI) algorithm.
- The MV-TSG faces challenges with the non-additive and non-Markovian variance metric, as well as non-stationary environment due to simultaneous policy updates.
- The MV-MAPI algorithm converges to a first-order stationary point, with specific conditions for local Nash equilibria and local optima.
- To solve large-scale MV-TSGs with unknown environmental parameters, a multi-agent reinforcement learning algorithm named MV-MATRPO is proposed.
Read Full Article
21 Likes
Arxiv
3d
56
Image Credit: Arxiv
Patronus: Bringing Transparency to Diffusion Models with Prototypes
- Diffusion-based generative models, such as Denoising Diffusion Probabilistic Models (DDPMs), have achieved remarkable success in image generation.
- A new interpretable diffusion model called Patronus is introduced, which integrates a prototypical network into DDPMs.
- Patronus enhances interpretability by showing the learned prototypes and how they influence the generation process.
- The model supports downstream tasks like image manipulation and can reveal shortcut learning in the generation process.
Read Full Article
3 Likes
Arxiv
3d
360
Image Credit: Arxiv
Quantum Doeblin Coefficients: Interpretations and Applications
- This study focuses on investigating quantum Doeblin coefficients as a generalization of classical concepts in information theory.
- The researchers define new quantum Doeblin coefficients with desirable properties and efficient computability.
- Various interpretations of the quantum Doeblin coefficients are presented, including their representations as minimal singlet fractions and exclusion values.
- The study also explores multiple applications of quantum Doeblin coefficients in various areas, providing improvements over prior literature in terms of generality and efficiency.
Read Full Article
21 Likes
Arxiv
3d
384
Image Credit: Arxiv
Nonhuman Primate Brain Tissue Segmentation Using a Transfer Learning Approach
- Accurate brain tissue segmentation in non-human primates (NHPs) is critical for understanding neurological disorders.
- A novel approach utilizing STU-Net with transfer learning enhances segmentation accuracy in NHP brain MRI.
- The method effectively delineates complex tissue boundaries and captures fine anatomical details specific to NHP brains.
- This study introduces a robust method for multi-class brain tissue segmentation in NHPs, benefiting research in evolutionary neuroscience and preclinical studies of neurological disorders.
Read Full Article
23 Likes
Arxiv
3d
204
Image Credit: Arxiv
Nested Stochastic Gradient Descent for (Generalized) Sinkhorn Distance-Regularized Distributionally Robust Optimization
- The paper proposes a nested stochastic gradient descent algorithm for solving regularized nonconvex Distributionally Robust Optimization (DRO) problems.
- The algorithm is designed to handle DRO problems with generalized Sinkhorn distance and nonconvex, unbounded loss functions.
- The proposed algorithm has polynomial iteration and sample complexities that are independent of data size and parameter dimension.
- Numerical experiments demonstrate the efficiency and robustness of the algorithm.
Read Full Article
12 Likes
Arxiv
3d
44
Image Credit: Arxiv
Identifying Multi-modal Knowledge Neurons in Pretrained Transformers via Two-stage Filtering
- Recent advances in large language models (LLMs) have led to the development of multimodal LLMs (MLLMs) in the fields of natural language processing (NLP) and computer vision.
- A study proposes a method to identify neurons associated with specific knowledge in MiniGPT-4, a Transformer-based multimodal LLM.
- The method involves two stages: activation differences filtering using inpainting and gradient-based filtering using GradCAM.
- Experiments show that the proposed method can locate knowledge with higher accuracy and contributes to the visualization and explainability of knowledge in MLLMs.
Read Full Article
2 Likes
Arxiv
3d
72
Image Credit: Arxiv
SUV: Scalable Large Language Model Copyright Compliance with Regularized Selective Unlearning
- Researchers propose the SUV framework to prevent Large Language Models (LLMs) from memorizing copyrighted content while preserving its overall utility.
- SUV constructs a dataset capturing instances of copyrighted infringement cases and unlearns the content from LLMs using Direct Preference Optimization (DPO).
- To mitigate the degradation in LLMs' performance on unrelated tasks, SUV integrates gradient projection and Fisher information regularization.
- Experiments on a large-scale dataset of 500 copyrighted books demonstrate the scalability and efficacy of SUV in reducing verbatim memorization without significant impact on unrelated tasks.
Read Full Article
4 Likes
Arxiv
3d
24
Image Credit: Arxiv
XL-Instruct: Synthetic Data for Cross-Lingual Open-Ended Generation
- Cross-lingual open-ended generation is an important yet understudied problem.
- XL-AlpacaEval is a new benchmark for evaluating cross-lingual generation capabilities in Large Language Models (LLMs).
- XL-Instruct is a high-quality synthetic data generation method that significantly improves model performance.
- XL-Instruct shows strong zero-shot transfer to both English-only and multilingual generation tasks.
Read Full Article
1 Like
Arxiv
3d
256
Image Credit: Arxiv
Engineering Microbial Symbiosis for Mars Habitability
- Recent advancements in synthetic biology and genetic engineering offer opportunities for addressing challenges in colonizing Mars.
- The paper examines the potential for creating symbiotic relationships between terrestrial microbes and hypothetical Martian life forms.
- Methods to engineer life forms capable of enduring Martian conditions are proposed, inspired by natural examples of endosymbiosis.
- The ethical, political, and technological challenges of introducing engineered life to Mars are critically evaluated, with an emphasis on international collaboration and planetary protection policies.
Read Full Article
15 Likes
Arxiv
3d
232
Image Credit: Arxiv
VLM-C4L: Continual Core Dataset Learning with Corner Case Optimization via Vision-Language Models for Autonomous Driving
- Handling complex environments in autonomous driving is a challenge due to the scarcity and diversity of extreme scenario datasets.
- Current autonomous driving models struggle to manage corner cases, posing a significant safety risk.
- VLM-C4L is a continual learning framework that enhances corner case datasets using Vision-Language Models (VLMs).
- VLM-C4L enables incremental learning from diverse corner cases while preserving performance on routine scenarios.
Read Full Article
13 Likes
Arxiv
3d
204
Image Credit: Arxiv
Concorde: Fast and Accurate CPU Performance Modeling with Compositional Analytical-ML Fusion
- Researchers have developed Concorde, a methodology for learning fast and accurate performance models of microarchitectures.
- Concorde uses compact performance distributions to predict program behavior based on different microarchitectural components.
- Experiments show that Concorde is over five orders of magnitude faster than a reference cycle-level simulator.
- It has an average Cycles-Per-Instruction (CPI) prediction error of about 2% across various benchmarks.
Read Full Article
12 Likes
For uninterrupted reading, download the app