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

A naukri.com initiative

New

Home

ML News

Arxiv

Image Credit: Arxiv

Asymptotic Theory for IV-Based Reinforcement Learning with Potential Endogeneity

Researchers propose a class of instrument variable-based reinforcement learning (IV-RL) algorithms to address reinforcement bias in data analysis.
The interaction between data generation and data analysis leads to reinforcement bias, exacerbating the endogeneity problem.
The proposed IV-RL algorithms are incorporated into a stochastic approximation framework and have theoretical properties.
The analysis also includes formulas for inference on optimal policies and highlights how intertemporal dependencies affect inference.

Read Full Article

Arxiv

326

Image Credit: Arxiv

Optimizing Convolutional Neural Networks for Chronic Obstructive Pulmonary Disease Detection in Clinical Computed Tomography Imaging

Researchers aimed to optimize the detection of Chronic Obstructive Pulmonary Disease (COPD) using convolutional neural networks (CNN).
They explored manually adjusted versus automated window-setting optimization (WSO) on computed tomography (CT) images.
The study utilized 7,194 CT images and compared various CNN architectures.
By applying automated WSO, the DenseNet model achieved a mean AUC of 0.82 for detecting COPD.

Read Full Article

19 Likes

Arxiv

Image Credit: Arxiv

Measurement-based quantum computation from Clifford quantum cellular automata

Measurement-based quantum computation (MBQC) is a paradigm for quantum computation driven by local measurements on an entangled resource state.
In this work, it is shown that MBQC is related to a model of quantum computation called Clifford quantum cellular automata (CQCA).
Certain MBQCs can be constructed directly from CQCAs, providing a simple and intuitive circuit model representation of MBQC based on CQCA.
MBQC-based Ansätze for parameterized quantum circuits are constructed, which can have significantly different performances on various learning tasks.

Read Full Article

1 Like

Arxiv

312

Image Credit: Arxiv

AgentBoard: An Analytical Evaluation Board of Multi-turn LLM Agents

AgentBoard is a comprehensive benchmark and evaluation framework for analyzing Large Language Models (LLMs) as general-purpose agents.
The framework addresses challenges in benchmarking agent performance across diverse scenarios and ensuring multi-round interactions.
AgentBoard introduces a fine-grained progress rate metric and a comprehensive evaluation toolkit for multi-faceted analysis.
The framework aims to demystify agent behaviors and accelerate the development of stronger LLM agents.

Read Full Article

18 Likes

Discover more

Arxiv

Image Credit: Arxiv

ProCNS: Progressive Prototype Calibration and Noise Suppression for Weakly-Supervised Medical Image Segmentation

ProCNS is a novel weakly-supervised segmentation approach for medical image segmentation.
It addresses the challenge of generating accurate pseudo-labels in noisy regions.
ProCNS consists of two modules: Prototype-based Regional Spatial Affinity (PRSA) and Adaptive Noise Perception and Masking (ANPM).
Experimental results show ProCNS outperforms existing methods in six medical image segmentation tasks.

Read Full Article

4 Likes

Arxiv

Image Credit: Arxiv

Learning Mutual Excitation for Hand-to-Hand and Human-to-Human Interaction Recognition

Recognizing interactive actions, including hand-to-hand interaction and human-to-human interaction, has attracted increasing attention.
A mutual excitation graph convolutional network (me-GCN) is proposed to model the semantic relationships between entities.
me-GC utilizes a mutual topology excitation module to model the mutual constraints between individual entities.
Experimental results demonstrate the superiority of me-GCN over existing GCN-based and Transformer-based methods.

Read Full Article

3 Likes

Arxiv

302

Image Credit: Arxiv

A Phase Transition in Diffusion Models Reveals the Hierarchical Nature of Data

Diffusion models reveal the hierarchical and combinatorial nature of data.
The backward diffusion process exhibits a phase transition at a threshold time.
Reconstruction of high-level features drops suddenly after the transition.
Generative models can effectively model combinatorial data properties.

Read Full Article

18 Likes

Arxiv

156

Image Credit: Arxiv

DelGrad: Exact event-based gradients in spiking networks for training delays and weights

Spiking neural networks (SNNs) rely on the timing of signals for representing and processing information.
A new method called DelGrad has been proposed to compute exact loss gradients for both synaptic weights and delays in SNNs.
DelGrad eliminates the need for tracking additional variables and offers higher precision, efficiency, and suitability for neuromorphic hardware.
Experimental results demonstrate the memory efficiency, accuracy benefits, and potential for stabilizing networks against noise in SNNs trained using DelGrad on a neuromorphic platform.

Read Full Article

9 Likes

Arxiv

196

Image Credit: Arxiv

Future Events as Backdoor Triggers: Investigating Temporal Vulnerabilities in LLMs

Large language models (LLMs) can distinguish past from future events with 90% accuracy.
Backdoors triggered by a temporal distributional shift can activate when exposed to news headlines beyond their training cut-off dates.
Fine-tuning on helpful, harmless, and honest (HHH) data is effective in removing backdoor triggers in backdoored models.
Standard safety measures are enough to remove backdoors in models at the modest scale tested.

Read Full Article

11 Likes

Arxiv

Image Credit: Arxiv

Principal Component Flow Map Learning of PDEs from Incomplete, Limited, and Noisy Data

Researchers have developed a computational technique for modeling the evolution of dynamical systems in a reduced basis.
The focus of the study is on modeling partially-observed partial differential equations (PDEs) on high-dimensional non-uniform grids.
The technique addresses the limitations of previous work by considering noisy and limited data, simulating real-world data collection scenarios.
By leveraging recent advancements in PDE modeling, the researchers propose a neural network structure that is suitable for modeling PDEs with noisy and limited data.

Read Full Article

4 Likes

Arxiv

233

Image Credit: Arxiv

Multi-Source EEG Emotion Recognition via Dynamic Contrastive Domain Adaptation

Researchers propose a multi-source dynamic contrastive domain adaptation method (MS-DCDA) for EEG emotion recognition.
The method leverages domain knowledge from multiple sources and uses dynamically weighted learning for optimal tradeoff between domain transferability and discriminability.
The proposed MS-DCDA model achieves high accuracies in cross-subject and cross-session experiments on SEED and SEED-IV datasets.
Insights from the study suggest greater emotional sensitivity in frontal and parietal brain lobes, with potential implications for mental health interventions and personalized medicine.

Read Full Article

14 Likes

Arxiv

226

Image Credit: Arxiv

SpikingSSMs: Learning Long Sequences with Sparse and Parallel Spiking State Space Models

Spiking neural networks (SNNs) have gained attention for their low energy consumption and temporal dynamics.
Researchers have developed spiking state space models (SpikingSSMs) for long sequence learning.
SpikingSSMs integrate neuronal dynamics with state space models and utilize sparse synaptic computation.
The proposed SpikingSSM shows competitive performance on benchmark tasks and has potential as a low computation cost architecture for language models.

Read Full Article

13 Likes

Arxiv

296

Image Credit: Arxiv

Unlocking Global Optimality in Bilevel Optimization: A Pilot Study

Bilevel optimization is important in AI applications, but obtaining global optimality is challenging.
Bilevel problems often lack a benign landscape and may have multiple local solutions.
This paper explores global convergence theory for bilevel optimization.
Two sufficient conditions for global convergence are presented, with proofs and experimental validation.

Read Full Article

17 Likes

Arxiv

173

Image Credit: Arxiv

PIETRA: Physics-Informed Evidential Learning for Traversing Out-of-Distribution Terrain

Self-supervised learning is used for developing traversability models for off-road navigation.
Existing methods utilize evidential deep learning to quantify model uncertainty.
PIETRA is a self-supervised learning framework that integrates physics priors into evidential neural networks.
PIETRA improves learning accuracy and navigation performance in environments with distribution shifts.

Read Full Article

10 Likes

Arxiv

242

Image Credit: Arxiv

Data-driven decision-making under uncertainty with entropic risk measure

The entropic risk measure is commonly used in high-stakes decision making to account for uncertain losses.
An empirical entropic risk estimator is often biased and underestimates the true risk with limited data.
A bootstrapping procedure is proposed to debias the empirical entropic risk estimator, improving risk estimation.
The approach is applied to distributionally robust entropic risk minimization and insurance contract design problems.

Read Full Article

14 Likes

For uninterrupted reading, download the app