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

A naukri.com initiative

New

Home

ML News

Arxiv

190

Image Credit: Arxiv

Improving Equivariant Networks with Probabilistic Symmetry Breaking

Equivariant networks encode known symmetries into neural networks, but they are unable to break symmetries.
Equivariant networks must have at least the same self-symmetries as the input, which limits their ability to handle prediction tasks and generative models.
To address this limitation, equivariant conditional distributions are considered instead of equivariant functions.
The SymPE method, which uses symmetry-breaking positional encodings, allows the breaking of symmetries while retaining the inductive bias of symmetry in equivariant networks.

Read Full Article

11 Likes

Arxiv

103

Image Credit: Arxiv

Arch-LLM: Taming LLMs for Neural Architecture Generation via Unsupervised Discrete Representation Learning

Unsupervised representation learning is essential for applications like Neural Architecture Search (NAS).
Variational Autoencoders (VAEs) often result in a high percentage of invalid or duplicate architectures when sampling from the continuous representation space.
A Vector Quantized Variational Autoencoder (VQ-VAE) is introduced to learn a discrete latent space for neural architectures.
The VQ-VAE approach significantly improves the generation of valid and unique architectures in NAS.

Read Full Article

6 Likes

Arxiv

238

Image Credit: Arxiv

Concise One-Layer Transformers Can Do Function Evaluation (Sometimes)

Concise 1-layer transformers have the ability to evaluate arbitrary functions under certain input representations.
However, they are incapable of performing this task when the function's inputs and values are assigned to different input positions.
Concise 2-layer transformers can successfully evaluate functions even with challenging input representations.
Experimental findings suggest a correlation between what concise transformers can compute and what can be effectively learned.

Read Full Article

14 Likes

Arxiv

Image Credit: Arxiv

Few-Shot Graph Out-of-Distribution Detection with LLMs

Existing methods for graph out-of-distribution (OOD) detection rely on labeled in-distribution (ID) data.
A new approach, LLM-GOOD, combines large language models (LLMs) and graph neural networks (GNNs) to enhance data efficiency in graph OOD detection.
LLM-GOOD leverages LLMs to filter out likely OOD nodes, reducing human annotation burden.
Experiments show that LLM-GOOD reduces human annotation costs and outperforms state-of-the-art baselines in ID classification accuracy and OOD detection performance.

Read Full Article

5 Likes

Discover more

Arxiv

146

Image Credit: Arxiv

Estimating City-wide operating mode Distribution of Light-Duty Vehicles: A Neural Network-based Approach

A modular neural network (NN)-based framework has been proposed to estimate operating mode distributions of light-duty vehicles without relying on predefined driving cycles.
The method utilizes macroscopic variables such as speed, flow, and link infrastructure attributes to estimate operating modes like braking, idling, and cruising.
The proposed framework outperforms the Motor Vehicle Emission Simulator (MOVES) in calculating the operating mode distribution, achieving a closer match to actual operating mode distribution derived from trajectory data.
The average error in emission estimation across pollutants is 8.57% for the proposed method, lower than the 32.86% error for MOVES. CO2 estimation has an error of just 4% compared to 35% for MOVES.

Read Full Article

8 Likes

Arxiv

150

Image Credit: Arxiv

Multimodal Machine Learning for Real Estate Appraisal: A Comprehensive Survey

Real estate appraisal is transitioning from manual to automated valuation, with a focus on multimodal machine learning.
Multimodal machine learning integrates diverse data sources to improve prediction accuracy and interpretability.
This survey provides a comprehensive review of research efforts in the real estate domain.
Future research directions include exploring multimodal complementarity and technology advancements.

Read Full Article

9 Likes

Arxiv

368

Image Credit: Arxiv

Long-Term Electricity Demand Prediction Using Non-negative Tensor Factorization and Genetic Algorithm-Driven Temporal Modeling

This study proposes a framework for long-term electricity demand prediction based solely on historical consumption data.
The method combines Non-negative Tensor Factorization (NTF) and a Genetic Algorithm to optimize the hyperparameters of time series models.
Experiments using real-world electricity data from Japan show that the proposed method achieves lower mean squared error than baseline approaches.
The framework offers an interpretable, flexible, and scalable approach to long-term electricity demand prediction.

Read Full Article

22 Likes

Arxiv

345

Image Credit: Arxiv

T-CIL: Temperature Scaling using Adversarial Perturbation for Calibration in Class-Incremental Learning

T-CIL is a novel temperature scaling approach for class-incremental learning without a validation set for old tasks.
It leverages adversarially perturbed exemplars from memory to improve model confidence calibration.
The key idea of T-CIL is to perturb exemplars more strongly for old tasks than for the new task based on feature distance.
T-CIL outperforms various baselines in terms of calibration and can be integrated with existing class-incremental learning techniques.

Read Full Article

20 Likes

Arxiv

353

Image Credit: Arxiv

Landscape of Thoughts: Visualizing the Reasoning Process of Large Language Models

A new visualization tool called 'Landscape of Thoughts' has been introduced to inspect the reasoning paths of large language models (LLMs) on multi-choice datasets.
The tool represents the states in a reasoning path as feature vectors and visualizes them using t-SNE in two-dimensional plots.
Qualitative and quantitative analysis with Landscape of Thoughts helps distinguish between strong and weak models, correct and incorrect answers, and highlights different reasoning patterns.
The tool can also be adapted to predict properties and has been showcased by adapting it to a lightweight verifier that evaluates the correctness of reasoning paths.

Read Full Article

21 Likes

Arxiv

357

Image Credit: Arxiv

Reasoning of Large Language Models over Knowledge Graphs with Super-Relations

Large language models (LLMs) have made progress in reasoning over knowledge graphs.
Current methods suffer from a high non-retrieval rate, reducing accuracy in answering questions based on these graphs.
To overcome this issue, the concept of super-relations is introduced, enabling both forward and backward reasoning by summarizing and connecting relational paths within the graph.
The proposed ReKnoS framework improves retrieval efficiency and reasoning performance, demonstrating superior performance over existing baselines.

Read Full Article

21 Likes

Arxiv

Image Credit: Arxiv

AdaRank: Adaptive Rank Pruning for Enhanced Model Merging

AdaRank is a model merging framework that adaptively selects beneficial singular directions of task vectors.
The reliance on manual rank selection in existing SVD-based techniques leads to cross-task interference and suboptimal performance.
AdaRank dynamically prunes singular components causing interference, achieving optimal information allocation to each task vector.
Empirical results demonstrate that AdaRank consistently outperforms existing methods, reducing the performance gap between fine-tuned models.

Read Full Article

5 Likes

Arxiv

253

Image Credit: Arxiv

Data-Free Universal Attack by Exploiting the Intrinsic Vulnerability of Deep Models

Deep neural networks (DNNs) are susceptible to Universal Adversarial Perturbations (UAPs) that can deceive a target model across a wide range of samples.
In this paper, a novel data-free method called Intrinsic UAP (IntriUAP) is proposed to attack deep models without using any image samples.
The vulnerability of deep models is predominantly influenced by the linear components, which are leveraged in IntriUAP to achieve highly competitive performance in attacking popular image classification deep models.
The method also demonstrates strong black-box attack performance even with limited access to the victim model's layers.

Read Full Article

15 Likes

Arxiv

361

Image Credit: Arxiv

Fuzzy Cluster-Aware Contrastive Clustering for Time Series

Researchers propose a new approach called fuzzy cluster-aware contrastive clustering (FCACC) for unsupervised time series learning
FCACC combines representation learning and clustering objectives to capture complex patterns in unlabeled time series data
The approach uses a three-view data augmentation strategy and a cluster-aware hard negative sample generation mechanism to improve feature extraction and discriminative ability
Experimental results demonstrate that FCACC outperforms selected baseline methods on 40 benchmark datasets

Read Full Article

21 Likes

Arxiv

372

Image Credit: Arxiv

Interpretable Deep Learning Paradigm for Airborne Transient Electromagnetic Inversion

Researchers propose a unified and interpretable deep learning inversion paradigm for analyzing airborne transient electromagnetic (ATEM) data.
Conventional methods for processing ATEM data have limitations in dealing with noise and achieving accurate inversion results.
The proposed approach involves disentangled representation learning to decompose noisy data into noise and signal factors.
The method demonstrates enhanced reliability and interpretability, leading to accurate reconstruction of subsurface electrical structure.

Read Full Article

22 Likes

Arxiv

Image Credit: Arxiv

DREMnet: An Interpretable Denoising Framework for Semi-Airborne Transient Electromagnetic Signal

The semi-airborne transient electromagnetic method (SATEM) is a useful surveying technique for challenging terrains.
DREMnet is an interpretable decoupled representation learning framework that enhances the denoising process of SATEM signals.
DREMnet disentangles data into content and context factors, leading to robust and interpretable denoising results in complex conditions.
Experimental results demonstrate that DREMnet outperforms existing techniques, improving the accuracy of identifying subsurface electrical structures.

Read Full Article

5 Likes

For uninterrupted reading, download the app