A naukri.com initiative
ML News
Arxiv
3d
376
Image Credit: Arxiv
Online Convex Optimization and Integral Quadratic Constraints: A new approach to regret analysis
- A new approach to regret analysis of online convex optimization algorithms is proposed.
- The approach leverages Integral Quadratic Constraints (IQCs) to derive a semi-definite program for regret guarantee.
- The concept of variational IQCs is introduced, which generalizes IQCs to time-varying monotone operators.
- The results do not require the assumption of gradient boundedness or a bounded feasible set.
Read Full Article
22 Likes
Arxiv
3d
380
Image Credit: Arxiv
Exploring GPT-4 for Robotic Agent Strategy with Real-Time State Feedback and a Reactive Behaviour Framework
- Researchers explore the use of GPT-4 on a humanoid robot in simulation and real-world environments.
- The study focuses on a large language model (LLM) driven behavior method for the robotic agent.
- The method addresses concerns around safety, task transitions, time horizons, and state feedback.
- Experiments show that the approach produces feasible outputs with smooth transitions and achieves user requests successfully.
Read Full Article
22 Likes
Arxiv
3d
124
Image Credit: Arxiv
Interpretable Machine Learning in Physics: A Review
- Machine learning is transforming scientific fields by enabling scientific discoveries beyond human capabilities.
- Interpretable machine learning allows experts to comprehend the concepts underlying predictions.
- It increases trust in black-box methods, helps reduce errors, and enhances human-AI collaboration.
- This review examines the role of interpretability in machine learning applied to physics.
Read Full Article
7 Likes
Arxiv
3d
248
Image Credit: Arxiv
A Constrained Multi-Agent Reinforcement Learning Approach to Autonomous Traffic Signal Control
- A Constrained Multi-Agent Reinforcement Learning Approach to Autonomous Traffic Signal Control
- Traffic congestion in modern cities is exacerbated by the limitations of traditional fixed-time traffic signal systems.
- Adaptive Traffic Signal Control (ATSC) algorithms dynamically adjust signal timing based on real-time traffic conditions.
- The proposed algorithm, Multi-Agent Proximal Policy Optimization with Lagrange Cost Estimator (MAPPO-LCE), integrates the Lagrange multipliers method to balance rewards and constraints, improving traffic signal control policies.
Read Full Article
14 Likes
Arxiv
3d
80
Image Credit: Arxiv
Learning a Single Index Model from Anisotropic Data with vanilla Stochastic Gradient Descent
- Learning a Single Index Model (SIM) from anisotropic Gaussian inputs using vanilla Stochastic Gradient Descent (SGD) is investigated.
- The impact of the covariance matrix on the learning dynamics and sample complexity is analyzed.
- Results show that vanilla SGD adapts to the data's covariance structure automatically.
- Upper and lower bounds on the sample complexity are derived based on the covariance matrix, not the input data dimension.
Read Full Article
4 Likes
Arxiv
3d
208
Image Credit: Arxiv
Scalable Geometric Learning with Correlation-Based Functional Brain Networks
- Researchers have developed a novel geometric framework that uses diffeomorphic transformations to embed correlation matrices into a Euclidean space.
- The proposed method improves computational speed and enhances accuracy compared to conventional manifold-based approaches.
- This framework has been applied to behavior score prediction, subject fingerprinting in resting-state fMRI, and hypothesis testing in electroencephalogram data.
- An open-source MATLAB toolbox is provided to facilitate broader adoption and advance the application of correlation geometry in functional brain network research.
Read Full Article
12 Likes
Arxiv
3d
196
Image Credit: Arxiv
MKA: Leveraging Cross-Lingual Consensus for Model Abstention
- Reliability of LLMs is questionable even as they get better at more tasks.
- This work focuses on utilizing the multilingual knowledge of an LLM to inform its decision to abstain or answer when prompted.
- A multilingual pipeline is developed to calibrate the model's confidence and let it abstain when uncertain.
- Results show significant improvement in accuracy, with a $71.2% improvement for Bengali and a 15.5% improvement for English.
Read Full Article
11 Likes
Arxiv
3d
104
Image Credit: Arxiv
Integral regularization PINNs for evolution equations
- Evolution equations, including ODEs and PDEs, are challenging to integrate accurately over long times.
- Physics-informed neural networks (PINNs) offer a mesh-free framework for solving PDEs but suffer from temporal error accumulation.
- To address this, integral regularization PINNs (IR-PINNs) introduce an integral-based residual term to improve temporal accuracy.
- IR-PINNs outperform original PINNs and other methods in capturing long-time behaviors, providing a robust and accurate solution for evolution equations.
Read Full Article
6 Likes
Arxiv
3d
340
Image Credit: Arxiv
Short-video Propagation Influence Rating: A New Real-world Dataset and A New Large Graph Model
- Short-video platforms have gained immense popularity, captivating the interest of millions, if not billions, of users globally.
- Researchers propose a new Short-video Propagation Influence Rating (SPIR) task and introduce a new Cross-platform Short-Video (XS-Video) dataset.
- The XS-Video dataset includes 117,720 videos, 381,926 samples, and 535 topics across 5 biggest Chinese platforms, annotated with the propagation influence from level 0 to 9.
- They also propose a Large Graph Model (LGM) named NetGPT to predict the long-term propagation influence of short-videos.
Read Full Article
20 Likes
Arxiv
3d
348
Image Credit: Arxiv
THEMIS: Towards Practical Intellectual Property Protection for Post-Deployment On-Device Deep Learning Models
- On-device deep learning (DL) has gained popularity in mobile apps for offline model inference and user privacy preservation.
- However, it introduces vulnerabilities related to model-stealing attacks and intellectual property infringement.
- A proposed solution called THEMIS lifts the read-only restriction of on-device DL models and protects the model owner's intellectual property.
- Experimental results and investigation of real-world DL mobile apps demonstrate the practicality and effectiveness of THEMIS.
Read Full Article
20 Likes
Arxiv
3d
308
Image Credit: Arxiv
Evaluation of (Un-)Supervised Machine Learning Methods for GNSS Interference Classification with Real-World Data Discrepancies
- The accuracy and reliability of vehicle localization on roads are crucial for applications such as self-driving cars, toll systems, and digital tachographs.
- Recent approaches based on machine learning (ML) have shown superior performance in monitoring interference, but their feasibility in real-world applications and environments has yet to be assessed.
- This study describes large-scale measurement campaigns conducted in real-world settings to evaluate supervised ML-based methods and their performance in real-world settings.
- The study also explores the challenges of combining datasets, outlier detection, and data augmentation techniques in adapting ML models to changes in the datasets.
Read Full Article
18 Likes
Arxiv
3d
152
Image Credit: Arxiv
Force-Free Molecular Dynamics Through Autoregressive Equivariant Networks
- Scientists introduce TrajCast, a transferable and data-efficient framework based on autoregressive equivariant message passing networks for molecular dynamics simulations.
- TrajCast allows for large-scale simulations over extended timescales, enabling the acceleration of materials discovery and exploration of physical phenomena beyond the reach of traditional simulations and experiments.
- Benchmarking of TrajCast demonstrates excellent agreement with reference MD simulations for structural, dynamical, and energetic properties across various systems.
- The open-source implementation of TrajCast is available on GitHub for accessibility and further development.
Read Full Article
9 Likes
Arxiv
3d
188
Image Credit: Arxiv
Adaptive Attention-Based Model for 5G Radio-based Outdoor Localization
- Radio-based localization in dynamic environments, such as urban and vehicular settings, requires systems that can efficiently adapt to varying signal conditions and environmental changes.
- This work presents an adaptive localization framework that combines shallow attention-based models with a router/switching mechanism based on a single-layer perceptron (SLP).
- The framework utilizes three low-complexity localization models optimized for different scenarios, allowing seamless adaptation to diverse deployment conditions.
- Real-world vehicle localization data collected from a massive MIMO base station (BS) validates the framework's ability to maintain high localization accuracy.
Read Full Article
11 Likes
Arxiv
3d
220
Image Credit: Arxiv
Free Parametrization of L2-bounded State Space Models
- Structured state-space models (SSMs) are widely used in machine learning and control.
- L2RU is a new parametrization of SSMs that ensures stability and robustness.
- L2RU enables unconstrained optimization using standard methods such as gradient descent.
- L2RU shows superior performance in system identification and control applications.
Read Full Article
13 Likes
Arxiv
3d
32
Image Credit: Arxiv
A Channel-Triggered Backdoor Attack on Wireless Semantic Image Reconstruction
- Researchers have proposed a novel attack paradigm, called Channel-Triggered Backdoor Attack (CT-BA), to exploit security vulnerabilities in wireless communication systems using deep learning.
- CT-BA leverages specific wireless channels as backdoor triggers, utilizing channel gain and channel noise as potential triggers.
- The attack demonstrates high success rates and effectiveness in various Semantic Communication (SemCom) systems and image reconstruction models.
- The study highlights the need for developing defense mechanisms to mitigate the risks associated with such backdoor attacks on wireless communication systems.
Read Full Article
1 Like
For uninterrupted reading, download the app