A naukri.com initiative
ML News
Arxiv
1w
87
Image Credit: Arxiv
LoLaFL: Low-Latency Federated Learning via Forward-only Propagation
- Federated Learning (FL) is a popular approach for edge learning with distributed data while ensuring privacy.
- The traditional FL with deep neural networks trained via backpropagation is not suitable for low-latency learning in 6G mobile networks.
- To address this, Low-Latency Federated Learning (LoLaFL) via forward-only propagation is proposed.
- LoLaFL enables layer-wise transmissions and aggregation, significantly reducing latency with two nonlinear aggregation schemes.
Read Full Article
5 Likes
Arxiv
1w
92
Image Credit: Arxiv
ReMoE: Fully Differentiable Mixture-of-Experts with ReLU Routing
- Sparsely activated Mixture-of-Experts (MoE) models are widely adopted to scale up model capacity without increasing the computation budget.
- ReMoE is a fully differentiable MoE architecture that offers a drop-in replacement for the conventional TopK+Softmax routing.
- ReMoE exhibits superior scalability with respect to the number of experts, surpassing traditional MoE architectures.
- The implementation of ReMoE based on Megatron-LM is available on GitHub.
Read Full Article
5 Likes
Arxiv
1w
79
Image Credit: Arxiv
Holistic Adversarially Robust Pruning
- Neural networks can be drastically shrunk in size by removing redundant parameters.
- Compression often leads to a drop in accuracy and lack of adversarial robustness.
- A new method called HARP copes with aggressive pruning better than previous approaches.
- HARP optimizes the compression rate and scoring connections for each layer individually, maintaining accuracy and robustness with a 99% reduction in network size.
Read Full Article
4 Likes
Arxiv
1w
67
Image Credit: Arxiv
Computing Gram Matrix for SMILES Strings using RDKFingerprint and Sinkhorn-Knopp Algorithm
- Researchers propose a kernel-based approach for encoding and analyzing molecular structures from SMILES strings.
- The approach involves computing a kernel matrix using the Sinkhorn-Knopp algorithm and kernel PCA for dimensionality reduction.
- null
- The method outperforms several baseline methods and shows potential for molecular design and drug discovery.
Read Full Article
4 Likes
Arxiv
1w
62
Image Credit: Arxiv
A Comprehensive Forecasting Framework based on Multi-Stage Hierarchical Forecasting Reconciliation and Adjustment
- A comprehensive demand forecasting system called "Multi-Stage Hierarchical Forecasting Reconciliation and Adjustment (Multi-Stage HiFoReAd)" has been introduced for Walmart's ad products.
- The system tackles hierarchical time series forecasting and addresses challenges of preserving seasonality, ensuring coherence, and improving accuracy.
- The system utilizes diverse models ensembled through Bayesian Optimization (BO) to achieve base forecasts.
- Experiments on Walmart's internal Ads-demand dataset and public datasets demonstrate significant improvement in error rates and coherence of forecasts at all hierarchical levels.
Read Full Article
3 Likes
Arxiv
1w
368
Image Credit: Arxiv
FROC: Building Fair ROC from a Trained Classifier
- This paper discusses the issue of fair probabilistic binary classification with binary protected groups.
- The objective is to design a fair classifier that is fair to both protected groups, irrespective of the threshold used by the practitioner.
- The proposed method, called FROC, introduces a threshold query model on ROC curves to transform a potentially unfair classifier's output to a fair classifier.
- The algorithm achieves the theoretical optimal guarantees and is evaluated on various real-world datasets.
Read Full Article
22 Likes
Arxiv
1w
255
Image Credit: Arxiv
Generative AI for Banks: Benchmarks and Algorithms for Synthetic Financial Transaction Data
- Generative AI may offer a solution to the challenges faced by the banking sector in using deep learning due to data sensitivity and regulatory constraints.
- This study evaluated five leading models - CTGAN, DGAN, Wasserstein GAN, FinDiff, and TVAE - for generating synthetic financial transaction data.
- While none of the algorithms can replicate the real data's graph structure, each excels in specific areas.
- DGAN is best for privacy-sensitive tasks, FinDiff and TVAE excel in data replication and augmentation, and CTGAN achieves a balance across all criteria.
Read Full Article
15 Likes
Arxiv
1w
108
Image Credit: Arxiv
A parametric algorithm is optimal for non-parametric regression of smooth functions
- A parametric algorithm called PADUA is introduced for the regression problem of smooth functions.
- PADUA aims to achieve a uniform error bound across the entire domain.
- It provides performance guarantees optimal up to constant or logarithmic factors.
- PADUA has optimal sample and space complexity, as demonstrated through numerical experiments.
Read Full Article
6 Likes
Arxiv
1w
385
Image Credit: Arxiv
Extending TWIG: Zero-Shot Predictive Hyperparameter Selection for KGEs based on Graph Structure
- Knowledge Graph Embeddings (KGEs) have been developed to analyze KGs and predict new facts based on the information in a KG.
- The Topologically-Weighted Intelligence Generation (TWIG) model is an extension of KGEs that can simulate the performance of KGE models on different hyperparameter settings and KGs.
- TWIG can accurately predict hyperparameter performance on unseen KGs in the zero-shot setting, suggesting the potential for pre-hoc hyperparameter selection using TWIG-like methods.
- Further research can explore the use of TWIG to determine optimal hyperparameter selection for KGE models without the need for a full hyperparameter search.
Read Full Article
23 Likes
Arxiv
1w
259
Image Credit: Arxiv
MARIA: a Multimodal Transformer Model for Incomplete Healthcare Data
- MARIA (Multimodal Attention Resilient to Incomplete datA) is a transformer-based deep learning model developed to address the challenge of managing missing data in healthcare applications.
- Unlike conventional approaches that rely on imputation, MARIA utilizes a masked self-attention mechanism to process only available data, without generating synthetic values.
- MARIA outperformed 10 state-of-the-art machine learning and deep learning models in 8 diagnostic and prognostic tasks, demonstrating its superior performance and resilience to varying levels of data incompleteness.
- This innovative model has the potential to significantly enhance comprehensive diagnostic and predictive models in critical healthcare applications.
Read Full Article
15 Likes
Arxiv
1w
402
Image Credit: Arxiv
Entropy Regularized Task Representation Learning for Offline Meta-Reinforcement Learning
- Offline meta-reinforcement learning aims to equip agents with the ability to rapidly adapt to new tasks by training on data from a set of different tasks.
- Context-based approaches suffer from distribution mismatch, limiting their ability to generalize to the test tasks.
- A new approach is proposed to minimize the mutual information between task representations and behavior policy, improving generalization ability.
- The approach outperforms prior methods in both in-distribution and out-of-distribution tasks in MuJoCo environments.
Read Full Article
24 Likes
Arxiv
1w
8
Image Credit: Arxiv
Hierarchical Subspaces of Policies for Continual Offline Reinforcement Learning
- Agents in dynamic domains like autonomous robotics and video game simulations must adapt to new tasks while retaining past knowledge.
- Continual Reinforcement Learning poses challenges of forgetting past knowledge and scalability.
- A novel framework called HIerarchical LOW-rank Subspaces of Policies (HILOW) is introduced for continual learning in offline navigation settings.
- HILOW leverages hierarchical policy subspaces for flexible and efficient adaptation to new tasks while preserving existing knowledge.
Read Full Article
Like
Arxiv
1w
29
Image Credit: Arxiv
Diffusion priors for Bayesian 3D reconstruction from incomplete measurements
- Many inverse problems require prior information to restrict the class of admissible models.
- Diffusion models are explored as priors in Bayesian 3D reconstruction.
- 3D point clouds are used to represent objects and incorporate incomplete and noisy data.
- Diffusion model priors enable 3D reconstruction from sparse and low-resolution observations.
Read Full Article
1 Like
Arxiv
1w
247
Image Credit: Arxiv
Robust Federated Learning in the Face of Covariate Shift: A Magnitude Pruning with Hybrid Regularization Framework for Enhanced Model Aggregation
- The development of highly sophisticated neural networks has allowed for fast progress in every field of computer vision, however, applications where annotated data is prohibited due to privacy or security concerns remain challenging.
- Federated Learning (FL) offers a promising framework for individuals aiming to collaboratively develop a shared model while preserving data privacy.
- A novel FL framework is proposed to mitigate the adverse effects of covariate shifts among federated clients by combining individual parameter pruning and regularization techniques to improve the robustness of individual clients' models to aggregate.
- Empirical findings substantiate the effectiveness of the proposed methodology across common benchmark datasets and in the presence of both high and low covariate shifts among client data distributions.
Read Full Article
14 Likes
Arxiv
1w
284
Image Credit: Arxiv
DroughtSet: Understanding Drought Through Spatial-Temporal Learning
- DroughtSet is a new dataset that integrates relevant predictive features and three drought indices from multiple remote sensing and reanalysis datasets across the United States.
- The dataset aims to provide the machine learning community with a benchmark to develop drought prediction models and time-series forecasting methods.
- The researchers also propose a spatial-temporal model, SPDrought, that learns from the spatial and temporal information of physical and biological features to predict three types of droughts simultaneously.
- The study provides insights into the predictability and sensitivity of droughts to biological and physical conditions, contributing to the understanding and mitigation of drought risks.
Read Full Article
17 Likes
For uninterrupted reading, download the app