A naukri.com initiative
ML News
Arxiv
3d
151
Image Credit: Arxiv
S$^2$DN: Learning to Denoise Unconvincing Knowledge for Inductive Knowledge Graph Completion
- Inductive Knowledge Graph Completion (KGC) aims to infer missing facts between newly emerged entities within knowledge graphs (KGs).
- A Semantic Structure-aware Denoising Network (S^2DN) is proposed for inductive KGC.
- S^2DN addresses the challenges of semantic inconsistencies and noisy interactions in KGs.
- Experimental results show that S^2DN outperforms state-of-the-art models in preserving semantic consistency and filtering out unreliable interactions.
Read Full Article
9 Likes
Arxiv
3d
127
Image Credit: Arxiv
Measuring Cross-Modal Interactions in Multimodal Models
- Integrating AI in healthcare can greatly improve patient care and system efficiency, but the lack of explainability in AI systems hinders their clinical adoption.
- Existing explainability methods for AI models in healthcare are limited to unimodal settings and fail to capture cross-modal interactions.
- This paper introduces InterSHAP, a cross-modal interaction score that quantifies the contributions of individual modalities and their interactions without approximations.
- InterSHAP accurately measures cross-modal interactions, handles multiple modalities, and provides detailed explanations for individual samples in multimodal medical datasets.
Read Full Article
7 Likes
Arxiv
3d
286
Image Credit: Arxiv
Improving Quantization-aware Training of Low-Precision Network via Block Replacement on Full-Precision Counterpart
- This paper proposes a general framework for improving the training of low-precision networks through block replacement on full-precision counterparts.
- The framework allows the low-precision network to be guided by the full-precision partner during training, addressing the limitations of direct training of low-precision networks.
- By generating intermediate mixed-precision models through block-by-block replacement, the integration of quantized low-precision blocks into full-precision networks is enabled.
- Experimental results demonstrate that the proposed method achieves state-of-the-art results for 4-, 3-, and 2-bit quantization on ImageNet and CIFAR-10.
Read Full Article
17 Likes
Arxiv
3d
55
Image Credit: Arxiv
MarkovType: A Markov Decision Process Strategy for Non-Invasive Brain-Computer Interfaces Typing Systems
- This work focuses on improving the Rapid Serial Visual Presentation (RSVP) typing task in Brain-Computer Interfaces (BCIs) using noninvasive electroencephalography (EEG).
- The proposed approach, MarkovType, incorporates a Partially Observable Markov Decision Process (POMDP) to achieve better accuracy in symbol classification while controlling the classification speed.
- MarkovType is the first work to formulate the RSVP typing task as a POMDP for recursive classification.
- Experiments show that MarkovType outperforms competitors, achieving a more accurate and balanced typing system.
Read Full Article
3 Likes
Arxiv
3d
51
Image Credit: Arxiv
Bayesian Optimization for Unknown Cost-Varying Variable Subsets with No-Regret Costs
- Bayesian Optimization (BO) is a method for optimizing expensive-to-evaluate black-box functions.
- Traditional BO assumes full control over query variables, but in real-world scenarios, controlling certain variables may incur costs.
- This problem is known as Bayesian Optimization with cost-varying variable subsets (BOCVS).
- A new algorithm is proposed for BOCVS with random and unknown costs, achieving sub-linear rate in quality and cost regret.
Read Full Article
3 Likes
Arxiv
3d
11
Image Credit: Arxiv
Self-supervised Spatial-Temporal Learner for Precipitation Nowcasting
- Nowcasting, the short-term prediction of weather, is essential for making timely and weather-dependent decisions.
- Precipitation nowcasting aims to predict precipitation at a local level within a 6-hour time frame.
- A novel model called SpaT-SparK, which combines self-supervised learning with spatial-temporal learning, has been developed for precipitation nowcasting.
- SpaT-SparK outperforms existing baseline supervised models, providing more accurate nowcasting predictions.
Read Full Article
Like
Arxiv
3d
215
Image Credit: Arxiv
CNN-LSTM Hybrid Deep Learning Model for Remaining Useful Life Estimation
- Remaining Useful Life (RUL) estimation is crucial in Predictive Maintenance applications.
- Traditional regression methods have struggled for high accuracy in this domain.
- A hybrid approach combining Convolutional Neural Networks (CNNs) with Long Short-Term Memory (LSTM) networks is proposed for RUL estimation.
- The hybrid CNN-LSTM model achieves the highest accuracy, outperforming other methods.
Read Full Article
12 Likes
Arxiv
3d
382
Image Credit: Arxiv
Choose Your Explanation: A Comparison of SHAP and GradCAM in Human Activity Recognition
- Explaining machine learning models using eXplainable AI (XAI) techniques has become essential in high-stakes domains like healthcare.
- A comparative analysis of Shapley Additive Explanations (SHAP) and Gradient-weighted Class Activation Mapping (GradCAM) methods in human activity recognition (HAR) is presented.
- The study evaluates these methods on skeleton-based data from real-world datasets and provides insights into their strengths, limitations, and differences.
- SHAP provides detailed feature attribution, while GradCAM delivers faster, spatially oriented explanations, making them complementary for different applications.
Read Full Article
23 Likes
Arxiv
3d
290
Image Credit: Arxiv
Fair Distributed Machine Learning with Imbalanced Data as a Stackelberg Evolutionary Game
- Decentralized learning allows training of deep learning algorithms without centralizing datasets, improving data privacy and operational efficiency.
- Data imbalances in distributed learning, especially in medical fields, pose challenges due to different patient populations and data collection practices.
- The paper proposes two algorithms, DSWM and ASWM, for setting weights of each node's contribution in the global model.
- The ASWM algorithm significantly improves the performance of underrepresented nodes by 2.713% in AUC, while nodes with larger datasets experience only a modest decrease of 0.441%.
Read Full Article
17 Likes
Arxiv
3d
47
Image Credit: Arxiv
Explainable AI for Multivariate Time Series Pattern Exploration: Latent Space Visual Analytics with Time Fusion Transformer and Variational Autoencoders in Power Grid Event Diagnosis
- Detecting and analyzing complex patterns in multivariate time-series data is crucial for decision-making in urban and environmental system operations.
- A new visual analytics framework is proposed to reduce complex patterns into lower-dimensional latent spaces and visualize them in 2D.
- The framework integrates Time Fusion Transformer (TFT) and Variational Autoencoders (VAEs) to enable intuitive exploration of complex multivariate temporal patterns.
- A case study on power grid signal data demonstrates the framework's ability to identify multi-label grid event signatures and evaluate model performance and efficiency.
Read Full Article
2 Likes
Arxiv
3d
163
Image Credit: Arxiv
EF-Net: A Deep Learning Approach Combining Word Embeddings and Feature Fusion for Patient Disposition Analysis
- Researchers have developed a prediction model for patient disposition using EF-Net.
- The model incorporates categorical and numerical features to achieve higher accuracy.
- The EF-Net model achieved an accuracy of 95.33%, while the Ensemble Model achieved 96%.
- The experiment shows that EF-Net outperforms existing works in accuracy, AUROC, and F1-Score.
Read Full Article
9 Likes
Arxiv
3d
39
Image Credit: Arxiv
FedGAT: A Privacy-Preserving Federated Approximation Algorithm for Graph Attention Networks
- Federated training methods have gained popularity for graph learning with applications including friendship graphs of social media sites and customer-merchant interaction graphs of huge online marketplaces.
- The graph is partitioned across clients due to privacy regulations, preventing clients from accessing information stored on other clients.
- Cross-client edges in the graph present a challenge to federated training methods as training a graph model at one client requires feature information of nodes on the other end of cross-client edges.
- The Federated Graph Attention Network (FedGAT) algorithm is introduced to approximate the behavior of Graph Attention Networks (GATs) for semi-supervised node classification with reduced communication overhead.
Read Full Article
2 Likes
Arxiv
3d
35
Image Credit: Arxiv
Offline Reinforcement Learning for LLM Multi-Step Reasoning
- Offline reinforcement learning (RL) is proposed to improve the multi-step reasoning ability of large language models (LLMs).
- The method called OREO (Offline Reasoning Optimization) jointly learns a policy model and value function by optimizing the soft Bellman Equation.
- OREO reduces the need to collect pairwise data and enables better credit assignment in multi-step reasoning tasks.
- Empirically, OREO surpasses existing offline learning methods on multi-step reasoning benchmarks.
Read Full Article
2 Likes
Arxiv
3d
278
Image Credit: Arxiv
Sum-of-Squares Programming for Ma-Trudinger-Wang Regularity of Optimal Transport Maps
- Researchers have developed a computational approach that uses Sum-of-Squares (SOS) programming to verify the non-negativity of the Ma-Trudinger-Wang (MTW) tensor associated with ground cost functions in optimal transport.
- The MTW tensor provides a measure of curvature in optimal transport and is crucial for establishing continuity in the Monge optimal transport map.
- The proposed approach not only provides certificates of non-negativity for the MTW tensor but also computes an inner approximation of the region where MTW non-negativity holds.
- The SOS programming method has been applied to various practical ground cost functions to approximate the regions of regularity in their corresponding optimal transport maps.
Read Full Article
16 Likes
Arxiv
3d
115
Image Credit: Arxiv
Investigating the importance of social vulnerability in opioid-related mortality across the United States
- The opioid crisis remains a critical public health challenge in the United States. Despite national efforts to reduce opioid prescribing rates, opioid overdose deaths have tripled. This study investigates the importance of social vulnerability factors in opioid-related mortality. It analyzes county-level data and identifies patterns that require further investigation. The study employs machine learning models to predict county-level opioid-related mortality rates, identifying the most important social vulnerability factors.
Read Full Article
6 Likes
For uninterrupted reading, download the app