A naukri.com initiative
ML News
Arxiv
2d
313
Image Credit: Arxiv
Assessing Foundation Models for Sea Ice Type Segmentation in Sentinel-1 SAR Imagery
- Accurate segmentation of sea ice types is crucial for mapping and operational forecasting in ice-covered waters.
- Deep learning methods often require extensive labeled datasets, which are time-consuming to create.
- This study evaluates ten remote sensing foundation models (FMs) for sea ice type segmentation using Sentinel-1 SAR imagery.
- Among the selected models, Prithvi-600M outperforms the baseline models, while CROMA achieves a similar performance in F1-score.
Read Full Article
18 Likes
Arxiv
2d
47
Image Credit: Arxiv
MixFunn: A Neural Network for Differential Equations with Improved Generalization and Interpretability
- MixFunn is a novel neural network architecture designed to solve differential equations with improved precision, interpretability, and generalization capability.
- The architecture includes the mixed-function neuron and the second-order neuron, which enhance the representational flexibility and expressive power of the network.
- MixFunn achieves comparable or superior results with significantly fewer parameters compared to conventional approaches.
- It has been successfully applied in solving differential equations in classical mechanics, quantum mechanics, and fluid dynamics, showing higher accuracy and improved generalization beyond the training domain.
Read Full Article
2 Likes
Arxiv
2d
174
Image Credit: Arxiv
Efficient Verified Machine Unlearning For Distillation
- Efficient Verified Machine Unlearning For Distillation
- Growing data privacy demands require efficient machine unlearning methods for removing the influence of specific training points.
- The PURGE framework integrates verified unlearning with distillation by using constituent mapping and an incremental multi-teacher strategy.
- PURGE reduces retraining overhead and achieves significant speed-ups in the unlearning process while maintaining student accuracy.
Read Full Article
10 Likes
Arxiv
2d
119
Image Credit: Arxiv
Niyama : Breaking the Silos of LLM Inference Serving
- Niyama is a QoS-driven inference serving system that enables efficient co-scheduling of diverse workloads on shared infrastructure.
- Existing LLM serving frameworks rely on siloed infrastructure, resulting in operational inefficiencies and over-provisioning.
- Niyama introduces fine-grained QoS classification and a dynamic chunking mechanism to improve serving capacity by 32% compared to current deployments.
- Under extreme load, Niyama reduces SLO violations by an order of magnitude compared to current strategies.
Read Full Article
7 Likes
Arxiv
2d
138
Image Credit: Arxiv
Benchmarking Ultra-Low-Power $\mu$NPUs
- Efficient on-device neural network (NN) inference has various advantages over cloud-based processing, including predictable latency, enhanced privacy, greater reliability, and reduced operating costs for vendors.
- This paper presents the first comparative evaluation and independent benchmarks of commercially-available microcontroller-scale neural processing units (µNPUs) designed for ultra-low-power applications.
- A model compilation framework is developed and open-sourced to enable consistent benchmarking of quantized models across diverse µNPU hardware. The benchmark includes factors such as model inference latency, power consumption, and memory overhead.
- The analysis reveals expected performance trends and surprising disparities between hardware specifications and actual performance, including unexpected scaling behaviors with increasing model complexity. This provides valuable insights for hardware designers and software developers in the rapidly evolving space of µNPUs.
Read Full Article
8 Likes
Arxiv
2d
11
Image Credit: Arxiv
Generative Latent Neural PDE Solver using Flow Matching
- Autoregressive next-step prediction models are widely used for building data-driven neural solvers for time-dependent partial differential equations (PDEs).
- A new approach proposes a latent diffusion model for PDE simulation, reducing computational costs.
- Using an autoencoder, different types of meshes are mapped onto a unified structured latent grid, enabling the capture of complex geometries.
- The proposed model outperforms deterministic baselines in accuracy and long-term stability, demonstrating the potential of diffusion-based approaches for robust data-driven PDE learning.
Read Full Article
Like
Arxiv
2d
241
Image Credit: Arxiv
Tropical Bisectors and Carlini-Wagner Attacks
- Using a tropical symmetric metric as an activation function in the last layer improves the robustness of CNNs against attacks.
- The decision boundary of a tropical CNN is defined by tropical bisectors.
- A refined version of the Carlini-Wagner attack is proposed, tailored for the tropical architecture.
- Computational experiments with MNIST and LeNet5 demonstrate the improved success rate of the attacks.
Read Full Article
14 Likes
Arxiv
2d
158
Image Credit: Arxiv
Mamba-3D as Masked Autoencoders for Accurate and Data-Efficient Analysis of Medical Ultrasound Videos
- Ultrasound videos are an important form of clinical imaging data for diagnostic analysis.
- E-ViM$^3$ is a data-efficient Vision Mamba network that enhances space-time correlations.
- Enclosure Global Tokens (EGT) capture and aggregate global features effectively.
- With limited labels, E-ViM$^3$ achieves competitive performance in semantic analysis tasks.
Read Full Article
9 Likes
Arxiv
2d
210
Image Credit: Arxiv
SeisRDT: Latent Diffusion Model Based On Representation Learning For Seismic Data Interpolation And Reconstruction
- A new latent diffusion model based on representation learning for seismic data interpolation and reconstruction has been proposed.
- Traditional seismic data reconstruction methods struggle to handle large-scale continuous missing traces.
- The proposed latent diffusion transformer utilizes representation learning to address complex and irregular missing situations in seismic data.
- Reconstruction experiments on field and synthetic datasets show that the method achieves higher accuracy and can handle various complex missing scenarios.
Read Full Article
12 Likes
Arxiv
2d
214
Image Credit: Arxiv
Binary AddiVortes: (Bayesian) Additive Voronoi Tessellations for Binary Classification with an application to Predicting Home Mortgage Application Outcomes
- The Additive Voronoi Tessellations (AddiVortes) model is extended for binary classification using a probit model with a latent variable formulation.
- The AddiVortes model outperforms random forests, BART, and other black-box regression models in most cases.
- A study using AddiVortes to predict mortgage approval likelihood shows its effectiveness in capturing complex relationships within the data.
- The model has the potential to improve decision-making in mortgage approval processes.
Read Full Article
12 Likes
Arxiv
2d
222
Image Credit: Arxiv
Architecture of Information
- The paper explores an approach to constructing energy landscapes of a formal neuron and multilayer artificial neural networks (ANNs).
- The study of informational and thermodynamic entropy in formal neuron and ANN models leads to the conclusion about the energetic nature of informational entropy.
- Modeling ANNs as energy systems makes it possible to interpret the structure of their internal energy as an internal model of the external world.
- The presented research makes it possible to formulate a formal definition of information in terms of the interaction processes between the internal and external energy of the system.
Read Full Article
13 Likes
Arxiv
2d
230
Image Credit: Arxiv
Meta-Representational Predictive Coding: Biomimetic Self-Supervised Learning
- Self-supervised learning has become increasingly important in machine intelligence.
- Meta-representational predictive coding (MPC) offers a neurobiologically plausible framework for self-supervised learning.
- MPC sidesteps the need for learning a generative model of sensory input by learning to predict representations of the input.
- MPC uses active inference to drive the learning of representations by sampling informative portions of the sensorium.
Read Full Article
13 Likes
Arxiv
2d
134
Image Credit: Arxiv
ELM: Ensemble of Language Models for Predicting Tumor Group from Pathology Reports
- ELM (Ensemble of Language Models) is a novel ensemble-based approach introduced to address the bottleneck in manually extracting data from unstructured pathology reports for tumor group assignment.
- ELM leverages both small language models (SLMs) and large language models (LLMs), utilizing six fine-tuned SLMs.
- ELM requires five-out-of-six agreement for tumor group classification, and disagreements are arbitrated by an LLM with a curated prompt.
- Evaluation shows that ELM achieves an average precision and recall of 0.94, outperforming other approaches and enhancing operational efficiencies in the British Columbia Cancer Registry.
Read Full Article
8 Likes
Arxiv
2d
142
Image Credit: Arxiv
Structured and sparse partial least squares coherence for multivariate cortico-muscular analysis
- A structured and sparse partial least squares coherence algorithm (ssPLSC) has been proposed for multivariate cortico-muscular analysis.
- The ssPLSC approach addresses challenges such as high dimensionality, limited sample sizes, and the need for interpretability and spatial structure.
- An efficient alternating iterative algorithm has been developed to solve the optimization problem in ssPLSC and its convergence has been proven experimentally.
- Experimental results have demonstrated that ssPLSC outperforms representative multivariate cortico-muscular fusion methods in scenarios with limited sample sizes and high noise levels, making it a transformative tool for evaluating corticospinal pathway integrity in neurological disorders.
Read Full Article
8 Likes
Arxiv
2d
170
Image Credit: Arxiv
Enhancing Predictive Accuracy in Tennis: Integrating Fuzzy Logic and CV-GRNN for Dynamic Match Outcome and Player Momentum Analysis
- A new approach to game prediction in professional tennis is introduced, combining a multi-level fuzzy evaluation model with a CV-GRNN model.
- Critical statistical indicators are identified using Principal Component Analysis and a two-tier fuzzy model is developed based on Wimbledon data.
- The study reveals strong correlations among momentum indicators, such as Player Win Streak and Score Difference, providing insights into players transitioning between losing and winning streaks.
- By incorporating 15 statistically significant indicators in the CV-GRNN model, the accuracy increases to 86.64% and the mean squared error decreases by 49.21%.
Read Full Article
10 Likes
For uninterrupted reading, download the app