A naukri.com initiative
ML News
Arxiv
3d
172
Image Credit: Arxiv
IMPACT: A Generic Semantic Loss for Multimodal Medical Image Registration
- This work introduces IMPACT, a generic semantic similarity metric for multimodal medical image registration.
- IMPACT compares deep learning-based features extracted from medical images without requiring task-specific training.
- The proposed metric offers significant advantages such as robustness, scalability, and efficiency.
- Evaluation on challenging registration tasks demonstrated improved anatomical alignment and increased robustness.
Read Full Article
10 Likes
Arxiv
3d
204
Image Credit: Arxiv
It's a (Blind) Match! Towards Vision-Language Correspondence without Parallel Data
- The platonic representation hypothesis suggests that vision and language embeddings become more homogeneous as model and dataset sizes increase.
- The study investigates the feasibility of matching vision and language embeddings in an unsupervised manner, without parallel data.
- A novel heuristic is introduced to solve the unsupervised matching problem, outperforming previous solvers.
- The analysis shows that vision and language representations can be matched without supervision, enabling embedding semantic knowledge into other modalities without annotation.
Read Full Article
12 Likes
Arxiv
3d
292
Image Credit: Arxiv
Graph Neural Network-Based Predictive Modeling for Robotic Plaster Printing
- This work proposes a Graph Neural Network (GNN) modeling approach for predicting the resulting surface in a particle-based fabrication process.
- The GNN model uses robotic arm trajectory features, printing process parameters, and a particle representation of the wall domain and end effector.
- It acts as a simulator of the printing process and aims to generate the robotic arm trajectory and optimize the printing parameters.
- The proposed model outperforms an existing benchmark model, demonstrating improved performance and error scaling.
Read Full Article
17 Likes
Arxiv
3d
176
Image Credit: Arxiv
Traffic Engineering in Large-scale Networks with Generalizable Graph Neural Networks
- Traffic engineering (TE) in large-scale computer networks can address the scalability issue of traditional TE algorithms.
- TELGEN is a novel TE algorithm that achieves superior generalizability across diverse network conditions.
- It transforms the problem of predicting the optimal TE solution into predicting the optimal TE algorithm.
- TELGEN achieved less than 3% optimality gap and reduced solving time by up to 84% compared to classical optimal solvers.
Read Full Article
10 Likes
Arxiv
3d
184
Image Credit: Arxiv
A Comparison of Parametric Dynamic Mode Decomposition Algorithms for Thermal-Hydraulics Applications
- This work compares different parametric Dynamic Mode Decomposition (DMD) algorithms for thermal-hydraulics applications.
- DMD is an equation-free technique used to learn linear models from time series datasets.
- The standard DMD formulation cannot handle parametric time series, requiring different linear models for each parameter realization.
- The study explores three different thermal-hydraulics problems to assess the advantages and shortcomings of the deployed algorithms.
Read Full Article
11 Likes
Arxiv
3d
324
Image Credit: Arxiv
MB-ORES: A Multi-Branch Object Reasoner for Visual Grounding in Remote Sensing
- A unified framework, MB-ORES, has been proposed to integrate object detection (OD) and visual grounding (VG) for remote sensing (RS) imagery.
- MB-ORES fine-tunes an open-set object detector, using referring expression data, to support OD and establish a prior for VG task in remote sensing.
- The model has a multi-branch network that generates task-aware proposals by integrating spatial, visual, and categorical features, and an object reasoning network that assigns probabilities to proposals for final referring object localization.
- MB-ORES demonstrates superior performance on OPT-RSVG and DIOR-RSVG datasets, outperforming existing methods while maintaining classical OD capabilities.
Read Full Article
19 Likes
Arxiv
3d
136
Image Credit: Arxiv
GPU-centric Communication Schemes for HPC and ML Applications
- Compute nodes on modern heterogeneous supercomputing systems consist of CPUs, GPUs, and high-speed network interconnects.
- Parallelization is a technique used for scalable simulation and deep learning workloads on these systems.
- Communication bottlenecks from distributed execution of parallel workloads can impact performance.
- This survey explores GPU-centric communication schemes that move control path from CPU to GPU.
Read Full Article
8 Likes
Arxiv
3d
236
Image Credit: Arxiv
Enhancing Image Resolution of Solar Magnetograms: A Latent Diffusion Model Approach
- The study focuses on enhancing the resolution of solar magnetograms taken by the Michelson Doppler Imager (MDI).
- A novel diffusion model approach called Latent Diffusion Model (LDM) is used to super-resolve the MDI magnetograms.
- The resolution of MDI observations is enhanced from 2"/pixel to 0.5"/pixel to match the higher-resolution capabilities of the Helioseismic and Magnetic Imager (HMI).
- Reconstructed images are evaluated using classical metrics and physical properties are examined to ensure preservation.
Read Full Article
14 Likes
Arxiv
3d
240
Image Credit: Arxiv
Learning Velocity and Acceleration: Self-Supervised Motion Consistency for Pedestrian Trajectory Prediction
- Understanding human motion is crucial for accurate pedestrian trajectory prediction.
- This work proposes a self-supervised pedestrian trajectory prediction framework that explicitly models position, velocity, and acceleration.
- The model leverages velocity and acceleration information to enhance position prediction through feature injection and a self-supervised motion consistency mechanism.
- Experiments on the ETH-UCY and Stanford Drone datasets show that the proposed method achieves state-of-the-art performance.
Read Full Article
14 Likes
Arxiv
3d
104
Image Credit: Arxiv
Fair Dynamic Spectrum Access via Fully Decentralized Multi-Agent Reinforcement Learning
- Decentralized wireless network with several source-destination pairs sharing limited frequency bands.
- Sources adapt band selection strategy without information sharing.
- Proposed Fair Share RL (FSRL) solution achieves fairness without coordination.
- FSRL can be up to 89.0% fairer compared to baseline RL algorithm.
Read Full Article
6 Likes
Arxiv
3d
8
Image Credit: Arxiv
Solving the Best Subset Selection Problem via Suboptimal Algorithms
- Best subset selection in linear regression is a nonconvex problem that is challenging to solve.
- Finding the global optimal solution via an exact optimization method may be impractical for high-dimensional problems.
- This study introduces a new suboptimal procedure for best subset selection in linear regression.
- Comparative experiments with synthetic and real data show the competitive performance of the new algorithm.
Read Full Article
Like
Arxiv
3d
288
Image Credit: Arxiv
Contextual Preference Collaborative Measure Framework Based on Belief System
- This article introduces a preference collaborative measure framework based on an updated belief system.
- The framework aims to reduce human intervention and improve the accuracy and efficiency of preference measures.
- It proposes algorithms to discover common preferences, update the belief system, and classify preference rules.
- Experimental results show that the proposed algorithms outperform state-of-the-art algorithms in most aspects.
Read Full Article
17 Likes
Arxiv
3d
356
Image Credit: Arxiv
Sim-and-Real Co-Training: A Simple Recipe for Vision-Based Robotic Manipulation
- Large real-world robot datasets hold potential to train generalist robot models.
- Co-training policy on a mixture of simulation and real-world datasets improves performance.
- Simulation data can enhance real-world task performance by an average of 38%.
- Research presents a simple recipe for utilizing simulation data for vision-based robotic manipulation.
Read Full Article
21 Likes
Arxiv
3d
80
Image Credit: Arxiv
Policy Gradient for LQR with Domain Randomization
- Domain randomization (DR) enables sim-to-real transfer by training controllers on a distribution of simulated environments.
- Simple policy gradient (PG) methods are often used to solve DR, but the theoretical guarantees are limited.
- A convergence analysis of PG methods for domain-randomized linear quadratic regulation (LQR) is provided in this study.
- The study shows that PG converges globally under suitable bounds on the heterogeneity of sampled systems, and proposes a discount-factor annealing algorithm.
Read Full Article
4 Likes
Arxiv
3d
100
Image Credit: Arxiv
Exploring the Effect of Reinforcement Learning on Video Understanding: Insights from SEED-Bench-R1
- Recent advancements in Chain of Thought (COT) generation have improved the reasoning capabilities of Large Language Models (LLMs).
- SEED-Bench-R1 is a benchmark designed to evaluate post-training methods for Multimodal Large Language Models (MLLMs) in video understanding.
- Reinforcement Learning (RL) shows data efficiency and superior performance on both in-distribution and out-of-distribution tasks compared to supervised fine-tuning (SFT).
- However, RL often produces less logically coherent reasoning chains and has limitations such as inconsistent reasoning and overlooked visual cues.
Read Full Article
6 Likes
For uninterrupted reading, download the app