A naukri.com initiative
ML News
Arxiv
1d
237
Image Credit: Arxiv
MARIOH: Multiplicity-Aware Hypergraph Reconstruction
- MARIOH is a supervised approach for reconstructing the original hypergraph from its projected graph by leveraging edge multiplicity.
- The approach integrates several key ideas to efficiently explore the search space.
- MARIOH achieves up to 74.51% higher reconstruction accuracy compared to state-of-the-art methods in experiments using 10 real-world datasets.
Read Full Article
14 Likes
Arxiv
1d
210
Image Credit: Arxiv
Efficient Annotator Reliablity Assessment with EffiARA
- EffiARA is an annotation framework that supports the entire annotation pipeline for document-level annotation tasks.
- It provides insights into the reliability of individual annotators and the annotated dataset as a whole.
- EffiARA has been shown to improve classification performance and increase agreement among annotators.
- The EffiARA Python package and its webtool provide an accessible graphical user interface for the system.
Read Full Article
12 Likes
Arxiv
1d
327
Image Credit: Arxiv
Near Field Localization via AI-Aided Subspace Methods
- The adoption of extremely large antennas in high-frequency bands is driving the demand for accurate near-field localization.
- This research proposes AI-aided subspace methods for near-field localization, addressing limitations of conventional techniques.
- Specifically, NF-SubspaceNet uses deep learning to improve localization in challenging conditions, while DCD-MUSIC decouples angle and range estimation to reduce complexity.
- Simulation results show that these methods outperform classical and existing deep-learning-based techniques for near-field localization.
Read Full Article
19 Likes
Arxiv
1d
259
Image Credit: Arxiv
Bi-Grid Reconstruction for Image Anomaly Detection
- Significant advancements have been made in image anomaly detection using un- and self-supervised methods.
- The paper introduces GRAD: Bi-Grid Reconstruction for Image Anomaly Detection.
- GRAD employs two continuous grids to enhance anomaly detection from normal and abnormal perspectives.
- Evaluations demonstrate significant performance improvements in fine-grained anomaly detection with GRAD.
Read Full Article
15 Likes
Arxiv
1d
372
Image Credit: Arxiv
Deep Learning Model Predictive Control for Deep Brain Stimulation in Parkinson's Disease
- Researchers have developed a data-driven Model Predictive Control (MPC) algorithm for deep brain stimulation (DBS) in the treatment of Parkinson's disease (PD).
- Closed-loop DBS (CLDBS) utilizes neural oscillations as a feedback signal, resulting in improved treatment outcomes and reduced side effects compared to open-loop DBS.
- The proposed algorithm uses a multi-step predictor based on input-convex neural networks to model the future evolution of beta oscillations, improving prediction accuracy and simplifying online computation.
- Through simulations and tests with PD patients, the algorithm achieved significant reductions in tracking error and control activity compared to existing CLDBS algorithms, offering a potential advancement in DBS treatment for PD and other diseases.
Read Full Article
22 Likes
Arxiv
1d
161
Image Credit: Arxiv
On Benchmarking Code LLMs for Android Malware Analysis
- Large Language Models (LLMs) have demonstrated strong capabilities in various code intelligence tasks.
- Cama is a benchmarking framework designed to evaluate the effectiveness of Code LLMs in Android malware analysis tasks.
- Cama specifies structured model outputs to support malicious function identification and malware purpose summarization.
- Experiments using Cama provide insights into how Code LLMs interpret decompiled code and their sensitivity to function renaming.
Read Full Article
9 Likes
Arxiv
1d
176
Image Credit: Arxiv
Command A: An Enterprise-Ready Large Language Model
- Command A is an enterprise-ready large language model optimized for real-world use cases.
- It is multilingual and supports 23 languages, making it suitable for global businesses.
- Command A uses a novel hybrid architecture, balancing efficiency with high-performance retrieval augmented generation (RAG) capabilities.
- The model has undergone decentralised training, self-refinement algorithms, and model merging techniques to achieve its advanced abilities.
Read Full Article
10 Likes
Arxiv
1d
105
Image Credit: Arxiv
Energy Weighted Learning Progress Guided Interleaved Multi-Task Learning
- Humans can continuously acquire new skills and knowledge by exploiting existing ones for improved learning, without forgetting them.
- 'Continual learning' in machine learning aims to learn new information while preserving the previously acquired knowledge.
- A new method is proposed that interleaves tasks based on their 'learning progress' and energy consumption, mimicking human task learning.
- Experiments conducted with a robot learning setting show that the proposed method achieves better performance and reduces energy consumption for learning tasks.
Read Full Article
6 Likes
Arxiv
1d
320
Image Credit: Arxiv
Communication-Efficient l_0 Penalized Least Square
- This paper introduces a communication-efficient penalized regression algorithm for high-dimensional sparse linear regression models with massive data.
- The algorithm, named CESDAR, leverages an optimized distributed system communication algorithm and introduces the communication-efficient surrogate likelihood framework to enhance privacy and data security.
- It achieves the same statistical accuracy as the global estimator while significantly reducing communication costs.
- Simulations and real data benchmarks experiments demonstrate the efficiency and accuracy of the CESDAR algorithm.
Read Full Article
19 Likes
Arxiv
1d
286
Image Credit: Arxiv
$C^2$AV-TSE: Context and Confidence-aware Audio Visual Target Speaker Extraction
- Audio-Visual Target Speaker Extraction (AV-TSE) aims to enhance auditory perception using visual cues.
- A model-agnostic strategy called Mask-And-Recover (MAR) is proposed to improve extraction quality by integrating contextual correlations.
- The Fine-grained Confidence Score (FCS) model is introduced to assess extraction quality and guide improvement on low-quality segments.
- The proposed model-agnostic training paradigm demonstrated consistent performance improvements across various metrics on the VoxCeleb2 dataset.
Read Full Article
17 Likes
Arxiv
1d
274
Image Credit: Arxiv
CellVTA: Enhancing Vision Foundation Models for Accurate Cell Segmentation and Classification
- CellVTA is a novel method that enhances the performance of vision foundation models for cell instance segmentation.
- It incorporates a CNN-based adapter module to extract high-resolution spatial information from input images and injects it into the Vision Transformer (ViT) through a cross-attention mechanism.
- CellVTA achieves excellent results with 0.538 mPQ on the CoNIC dataset and 0.506 mPQ on the PanNuke dataset, surpassing state-of-the-art cell segmentation methods.
- The code and models for CellVTA are publicly available on GitHub at https://github.com/JieZheng-ShanghaiTech/CellVTA.
Read Full Article
16 Likes
Arxiv
1d
282
Image Credit: Arxiv
FeatInsight: An Online ML Feature Management System on 4Paradigm Sage-Studio Platform
- Feature management is essential for many online machine learning applications.
- FeatInsight is a system that supports the entire feature lifecycle, including design, storage, computation, verification, and lineage management.
- FeatInsight has been deployed in over 100 real-world scenarios on 4Paradigm's Sage Studio platform, handling a large feature space and enabling rapid updates.
- FeatInsight enhances feature design efficiency and improves feature computation performance.
Read Full Article
16 Likes
Arxiv
1d
304
Image Credit: Arxiv
Spingarn's Method and Progressive Decoupling Beyond Elicitable Monotonicity
- Spingarn's method and progressive decoupling algorithm address inclusion problems involving the sum of an operator and the normal cone of a linear subspace.
- This paper introduces progressive decoupling+ which incorporates separate relaxation parameters for the linkage subspace and its orthogonal complement.
- The convergence of progressive decoupling+ is proven under conditions linking relaxation parameters to the nonmonotonicity of their respective subspaces.
- The analysis shows that Spingarn's method and standard progressive decoupling also extend beyond the elicitable monotone setting.
Read Full Article
18 Likes
Arxiv
1d
37
Image Credit: Arxiv
PRISM-0: A Predicate-Rich Scene Graph Generation Framework for Zero-Shot Open-Vocabulary Tasks
- PRISM-0 is a framework for zero-shot open-vocabulary Scene Graph Generation (SGG).
- PRISM-0 overcomes training bias and lack of predicate diversity in SGG tasks.
- It uses a bottom-up approach to capture diverse predicate predictions.
- PRISM-0 improves downstream tasks such as Image Captioning and Sentence-to-Graph Retrieval.
Read Full Article
2 Likes
Arxiv
1d
48
Image Credit: Arxiv
Logical perspectives on learning statistical objects
- The relationship between learnability of a 'base class' of functions on a set X and learnability of a class of statistical functions derived from the base class is considered.
- The study examines Probably Approximately Correct (PAC) learning and online learning in terms of sample complexity and combinatorial dimensions of the base class.
- Improved bounds on the sample complexity of learning for statistical classes are established by using techniques from model theory.
- The focus is on classes derived from logical formulas, and the article discusses the relationship between learnability and properties of the formula.
Read Full Article
2 Likes
For uninterrupted reading, download the app