A naukri.com initiative
ML News
Arxiv
22h
306
Image Credit: Arxiv
ComFairGNN: Community Fair Graph Neural Network
- Graph Neural Networks (GNNs) have become the leading approach for addressing graph analytical problems in various real-world scenarios.
- GNNs may produce biased predictions against certain demographic subgroups due to node attributes and neighbors surrounding a node.
- Current research on GNN fairness often uses oversimplified fairness evaluation metrics, resulting in misleading impressions of fairness.
- ComFairGNN is a novel framework designed to mitigate community-level bias in GNNs by employing a learnable coreset-based debiasing function.
Read Full Article
18 Likes
Arxiv
22h
78
Image Credit: Arxiv
ResKoopNet: Learning Koopman Representations for Complex Dynamics with Spectral Residuals
- ResKoopNet is a novel method for approximating Koopman operator spectral components.
- It explicitly minimizes the spectral residual to compute Koopman eigenpairs.
- ResKoopNet provides more precise and complete spectra compared to existing methods.
- Experiments show superior accuracy for high-dimensional dynamical systems.
Read Full Article
4 Likes
Arxiv
22h
37
Image Credit: Arxiv
Interpretable Steering of Large Language Models with Feature Guided Activation Additions
- Effective and reliable control over large language model (LLM) behavior is a significant challenge.
- The existing activation steering methods lack precision and interpretability in influencing model outputs.
- Feature Guided Activation Additions (FGAA) is a novel activation steering method that provides better steering effects with coherence of model outputs.
- FGAA outperforms existing steering methods of CAA, SAE decoder steering, and SAE-TS in steering tasks across various models.
Read Full Article
2 Likes
Arxiv
22h
126
Image Credit: Arxiv
ACT-JEPA: Novel Joint-Embedding Predictive Architecture for Efficient Policy Representation Learning
- Learning efficient representations for decision-making policies is a challenge in imitation learning (IL).
- Self-supervised learning (SSL) offers an alternative by allowing models to learn from diverse, unlabeled data, including failures.
- ACT-JEPA is a novel architecture that integrates IL and SSL to enhance policy representations.
- ACT-JEPA improves the quality of representations by learning temporal environment dynamics and effectively generalizes to action sequence prediction.
Read Full Article
7 Likes
Arxiv
22h
41
Image Credit: Arxiv
Active teacher selection for reinforcement learning from human feedback
- Reinforcement learning from human feedback (RLHF) enables machine learning systems to learn objectives from human feedback.
- The Hidden Utility Bandit (HUB) framework is proposed to model differences in teacher rationality, expertise, and costliness, formalizing the problem of learning from multiple teachers.
- The Active Teacher Selection (ATS) algorithm outperforms baseline algorithms by actively selecting when and which teacher to query.
- The HUB framework and ATS algorithm facilitate future research on active teacher selection for robust reward modeling.
Read Full Article
2 Likes
Arxiv
22h
261
Image Credit: Arxiv
Meta ControlNet: Enhancing Task Adaptation via Meta Learning
- Diffusion-based image synthesis using ControlNet attracted attention recently.
- Vanilla ControlNet requires extensive training of around 5000 steps for a single task.
- A novel Meta ControlNet method is introduced which significantly reduces learning steps to 1000 and achieves zero-shot adaptability in edge-based tasks.
- Meta ControlNet achieves control in only 100 finetuning steps in complex non-edge tasks, outperforming existing methods.
Read Full Article
15 Likes
Arxiv
22h
362
Image Credit: Arxiv
Making Large Language Models Better Knowledge Miners for Online Marketing with Progressive Prompting Augmentation
- The paper discusses the development of mobile economy and the importance of efficient matching between user preferences and marketing campaigns.
- It introduces PAIR, a progressive prompting augmented mining framework for constructing a marketing-oriented knowledge graph with Large Language Models (LLMs).
- PAIR addresses issues such as uncontrollable relation generation, insufficient prompting ability, and high deployment cost of LLMs.
- The effectiveness of the proposed framework is validated through extensive experiments and practical applications in audience targeting.
Read Full Article
21 Likes
Arxiv
22h
313
Image Credit: Arxiv
IR2: Information Regularization for Information Retrieval
- IR2, Information Regularization for Information Retrieval, is a technique for reducing overfitting during synthetic data generation in settings with limited training data.
- Experimental results indicate that IR2 outperforms previous synthetic query generation methods and reduces cost by up to 50%.
- Three different regularization methods at different stages of the query synthesis pipeline are explored, offering varying degrees of performance improvement.
- The code, prompts, and synthetic data for IR2 are available on GitHub.
Read Full Article
18 Likes
Arxiv
22h
111
Image Credit: Arxiv
Machine Vision-Based Assessment of Fall Color Changes and its Relationship with Leaf Nitrogen Concentration
- This study focuses on the machine vision-based assessment of fall color changes in apple trees and its correlation with leaf nitrogen concentration.
- An image dataset was collected using a ground vehicle-based stereovision sensor to quantify the change in leaf color over a five-week period in a commercial orchard.
- A custom-defined metric, the 'yellowness index,' was used to estimate the proportion of yellow leaves per canopy.
- The study found that the 'yellowness index' was able to capture the gradual color transition from green to yellow, and trees with lower leaf nitrogen showed an earlier color change compared to trees with higher nitrogen.
Read Full Article
6 Likes
Arxiv
22h
0
Image Credit: Arxiv
Medical Spoken Named Entity Recognition
- Spoken Named Entity Recognition (NER) aims to extract named entities from speech and categorise them into types like person, location, organization, etc.
- VietMed-NER is the first spoken NER dataset in the medical domain, and the largest spoken NER dataset in the world for the number of entity types.
- Baseline results using various state-of-the-art pre-trained models show that pre-trained multilingual models generally outperform monolingual models on reference text and ASR output.
- The dataset can be utilized for text NER in the medical domain in other languages by translating the transcripts.
Read Full Article
Like
Arxiv
22h
18
Image Credit: Arxiv
Automate Strategy Finding with LLM in Quant Investment
- A new framework leveraging Large Language Models (LLMs) and multi-agent architectures has been proposed for quantitative stock investment in portfolio management and alpha mining.
- The framework incorporates LLMs to generate diversified alphas and utilizes a multi-agent approach to dynamically evaluate market conditions.
- The first module of the framework extracts predictive signals by analyzing numerical data, research papers, and visual charts.
- Extensive experiments on the Chinese stock markets demonstrate that this framework outperforms state-of-the-art baselines, highlighting the potential of AI-driven approaches in enhancing quantitative investment strategies.
Read Full Article
1 Like
Arxiv
22h
324
Image Credit: Arxiv
Streamlined optical training of large-scale modern deep learning architectures with direct feedback alignment
- Researchers implement a versatile and scalable training algorithm, direct feedback alignment, on a hybrid electronic-photonic platform.
- An optical processing unit achieves large-scale random matrix multiplications at speeds up to 1500 TeraOPS.
- Optical training of modern deep learning architectures, including Transformers with over 1 billion parameters, is performed with good performance on various tasks.
- The hybrid opto-electronic approach shows potential for ultra-deep and wide neural networks, offering a promising way to extend the growth of artificial intelligence beyond traditional approaches.
Read Full Article
19 Likes
Arxiv
22h
354
Image Credit: Arxiv
Batch, match, and patch: low-rank approximations for score-based variational inference
- Black-box variational inference (BBVI) scales poorly for estimating a multivariate Gaussian approximation with a full covariance matrix in high-dimensional problems.
- The batch-and-match (BaM) framework extends score-based BBVI and addresses the challenge of expensive storage and estimation of covariance matrices.
- BaM uses specialized updates to match scores of the target density and its Gaussian approximation, instead of relying on stochastic gradient descent.
- By integrating the updates with a more compact parameterization, BaM introduces a patch that projects covariance matrices into a more efficiently parameterized family of diagonal plus low rank matrices.
Read Full Article
21 Likes
Arxiv
22h
332
Image Credit: Arxiv
Learning Koopman-based Stability Certificates for Unknown Nonlinear Systems
- Koopman operator theory has gained attention for identifying nonlinear systems.
- New framework proposed to learn vector field and Lyapunov functions for unknown nonlinear systems.
- Algorithmic framework utilizes limited data sampled at low frequency.
- Learned Lyapunov functions can be formally verified and provide less conservative estimates of the region of attraction.
Read Full Article
19 Likes
Arxiv
22h
291
Image Credit: Arxiv
Gaussian entropic optimal transport: Schr\"odinger bridges and the Sinkhorn algorithm
- Entropic optimal transport problems are regularized versions of optimal transport problems.
- The Sinkhorn algorithm is commonly used to solve these problems for finite spaces.
- A finite-dimensional recursive formulation of the Sinkhorn algorithm for Gaussian multivariate models is proposed.
- The algorithm is closely related to the Kalman filter and provides closed form expressions for entropic transport maps and Schrödinger bridges.
Read Full Article
17 Likes
For uninterrupted reading, download the app