Machine Learning (ML) Latest News and Trending articles from all top sources only on Techminis

A naukri.com initiative

New

Home

ML News

Medium

267

Image Credit: Medium

How to Detect Outliers in Data Preprocessing? A Guide for Data Scientists ️

As data scientists, it is important to detect and handle outliers in data preprocessing.
Two powerful methods to detect outliers are Z-score and Interquartile Range (IQR).
Z-score measures the number of standard deviations a data point is away from the mean.
A Z-score greater than 3 or less than -3 indicates an outlier.

Read Full Article

16 Likes

Mit

233

Image Credit: Mit

Researchers teach LLMs to solve complex planning challenges

MIT researchers have developed a framework to guide large language models (LLMs) in solving complex planning problems like a human.
The framework allows users to describe the problem in natural language without needing specific examples for training the LLM.
The model encodes the user's text prompt for efficient solving of planning challenges using optimization software.
During problem formulation, the LLM checks its work at intermediate steps to rectify errors and ensure accurate planning.
The framework showed an 85 percent success rate in solving challenges like minimizing warehouse robot travel distance.
It can be applied to tasks such as crew scheduling and factory machine time management.
The research introduces a smart assistant framework that finds optimal plans even for complex or unusual rules.
The framework, LLM-Based Formalized Programming (LLMFP), prompts LLMs to reason about problems and determine solutions.
LLMFP self-assesses the solution and corrects any errors in the problem formulation for an accurate final plan.
The framework achieved an average success rate between 83-87% in diverse planning problems across multiple LLMs.

Read Full Article

14 Likes

Arxiv

Image Credit: Arxiv

LayerCraft: Enhancing Text-to-Image Generation with CoT Reasoning and Layered Object Integration

LayerCraft is an automated framework that enhances text-to-image generation with CoT reasoning and layered object integration.
It leverages large language models (LLMs) as autonomous agents for structured procedural generation.
LayerCraft enables customization of objects within an image and supports narrative-driven creation with minimal effort.
The framework democratizes creative image creation by providing non-experts with intuitive, precise control over T2I generation.

Read Full Article

5 Likes

Arxiv

357

Image Credit: Arxiv

Revisiting the Relationship between Adversarial and Clean Training: Why Clean Training Can Make Adversarial Training Better

Adversarial training (AT) is an effective technique for enhancing adversarial robustness, but it usually comes at the cost of a decline in generalization ability.
Recent studies have attempted to use clean training to assist adversarial training, yet there are contradictions among the conclusions.
The knowledge combinations transferred from clean-trained models to adversarially-trained models can be divided into two categories: reducing the learning difficulty and providing correct guidance.
By leveraging clean training, the performance of advanced AT methods can be further improved, and the problem of generalization degradation faced by AT can be alleviated.

Read Full Article

21 Likes

Discover more

Arxiv

218

Image Credit: Arxiv

ModelRadar: Aspect-based Forecast Evaluation

Accurate evaluation of forecasting models is essential for ensuring reliable predictions.
Current practices for evaluating and comparing forecasting models focus on summarizing performance into a single score, which may not provide enough information about model behavior under varying conditions.
To address this limitation, ModelRadar is proposed as a framework for evaluating univariate time series forecasting models across multiple aspects, such as stationarity, presence of anomalies, or forecasting horizons.
Comparing 24 forecasting methods, including classical approaches and different machine learning algorithms, NHITS, a state-of-the-art neural network architecture, performs best overall, but its superiority varies depending on the forecasting conditions.

Read Full Article

13 Likes

Arxiv

301

Image Credit: Arxiv

CF-CAM: Gradient Perturbation Mitigation and Feature Stabilization for Reliable Interpretability

CF-CAM is a novel framework that addresses the challenge of neural network decision-making opacity in deep learning.
Existing Class Activation Mapping (CAM) techniques for visualizing model decisions have trade-offs related to gradient perturbations and computational overhead.
CF-CAM employs a hierarchical importance weighting strategy and clustering techniques to enhance robustness against gradient noise and preserve discriminative features.
Experimental results show that CF-CAM outperforms state-of-the-art CAM methods in interpretability and robustness, making it suitable for critical applications like medical diagnosis and autonomous driving.

Read Full Article

18 Likes

Arxiv

331

Image Credit: Arxiv

Integrating Quantum-Classical Attention in Patch Transformers for Enhanced Time Series Forecasting

QCAAPatchTF is a quantum attention network integrated with an advanced patch-based transformer for time series forecasting.
The model uses a quantum-classical hybrid self-attention mechanism to capture multivariate correlations across time points.
It achieves state-of-the-art performance in long-term and short-term forecasting, classification, and anomaly detection tasks.
QCAAPatchTF demonstrates high accuracy and efficiency on complex real-world datasets.

Read Full Article

19 Likes

Arxiv

Image Credit: Arxiv

Enhancing Time Series Forecasting with Fuzzy Attention-Integrated Transformers

This paper introduces FANTF (Fuzzy Attention Network-Based Transformers), a novel approach that integrates fuzzy logic with existing transformer architectures to advance time series forecasting, classification, and anomaly detection tasks.
FANTF leverages a proposed fuzzy attention mechanism incorporating fuzzy membership functions to handle uncertainty and imprecision in noisy and ambiguous time series data.
The FANTF approach enhances its ability to capture complex temporal dependencies and multivariate relationships by embedding fuzzy logic principles into the self-attention module of the existing transformer's architecture.
Experimental evaluations on real-world datasets show that FANTF significantly enhances the performance of forecasting, classification, and anomaly detection tasks over traditional transformer-based models.

Read Full Article

2 Likes

Arxiv

353

Image Credit: Arxiv

Times2D: Multi-Period Decomposition and Derivative Mapping for General Time Series Forecasting

Time series forecasting is important in various domains such as energy management and financial markets.
The Times2D method transforms 1D time series into 2D space to capture complex temporal variations.
It consists of a Periodic Decomposition Block, First and Second Derivative Heatmaps, and an Aggregation Forecasting Block.
Times2D achieves state-of-the-art performance in both short-term and long-term forecasting.

Read Full Article

21 Likes

Arxiv

Image Credit: Arxiv

EMForecaster: A Deep Learning Framework for Time Series Forecasting in Wireless Networks with Distribution-Free Uncertainty Quantification

A deep learning framework called EMForecaster has been developed for time series forecasting in wireless networks.
EMForecaster employs patching and reversible instance normalization and mixing operations for efficient feature extraction.
The framework includes a conformal prediction mechanism for uncertainty quantification of forecasts.
EMForecaster outperforms current state-of-the-art DL approaches and achieves a superior balance between prediction interval width and coverage.

Read Full Article

3 Likes

Arxiv

304

Image Credit: Arxiv

Lorentzian Graph Isomorphic Network

Researchers have introduced the Lorentzian Graph Isomorphic Network (LGIN), a graph neural network designed to operate in hyperbolic spaces.
LGN incorporates curvature-aware aggregation functions that preserve the Lorentzian metric tensor, enhancing graph representation learning.
Through extensive evaluation on benchmark datasets, LGIN consistently outperforms or matches state-of-the-art graph neural networks, demonstrating its robustness and efficacy.
LGIN extends the concept of a powerful graph neural network to Riemannian manifolds, paving the way for advancements in hyperbolic graph learning.

Read Full Article

18 Likes

Arxiv

286

Image Credit: Arxiv

MetaCLBench: Meta Continual Learning Benchmark on Resource-Constrained Edge Devices

MetaCLBench is an end-to-end Meta-CL benchmark framework for edge devices.
It evaluates six representative Meta-CL approaches using both image and audio datasets.
The study reveals computational and memory costs associated with Meta-CL methods on edge devices.
Practical guidelines are provided for implementing Meta-CL on resource-constrained environments.

Read Full Article

17 Likes

Arxiv

293

Image Credit: Arxiv

Discriminative Subspace Emersion from learning feature relevances across different populations

A new Discriminative Subspace Emersion (DSE) method has been proposed.
DSE extends subspace learning to a general relevance learning framework.
The method identifies the most relevant features in distinguishing a classification task across two populations.
DSE is shown to accurately identify a common subspace even with a high degree of overlap between classes.

Read Full Article

17 Likes

Arxiv

Image Credit: Arxiv

Are Domain Generalization Benchmarks with Accuracy on the Line Misspecified?

Spurious correlations can hinder robust decision-making and generalization out-of-distribution (OOD).
Naive in-distribution empirical risk minimizers often achieve the best OOD accuracy.
Many popular benchmarks evaluating robustness to spurious correlations are misspecified.
The need to reassess how robustness to spurious correlations is evaluated is highlighted.

Read Full Article

2 Likes

Arxiv

143

Image Credit: Arxiv

Identifying Sparsely Active Circuits Through Local Loss Landscape Decomposition

A new decomposition method called Local Loss Landscape Decomposition (L3D) has been introduced to better understand the circuits employed by models.
L3D identifies a set of low-rank subnetworks in the parameter space that can reconstruct the gradient of the loss between any sample's output and a reference output vector.
The method was successfully tested on progressively more challenging toy models, showing its ability to recover associated subnetworks.
L3D was applied to a transformer model and a convolutional neural network, demonstrating its potential to identify interpretable and relevant circuits in parameter space.

Read Full Article

8 Likes

For uninterrupted reading, download the app