menu
techminis

A naukri.com initiative

google-web-stories
Home

>

ML News

ML News

source image

Arxiv

2h

read

19

img
dot

Image Credit: Arxiv

Benchmarking Federated Machine Unlearning methods for Tabular Data

  • This paper focuses on benchmarking machine unlearning methods for tabular data within a federated learning (FL) setting.
  • The study explores unlearning at the feature and instance levels using machine learning models.
  • The benchmarking methodology evaluates various unlearning algorithms and compares their fidelity, certifiability, and computational efficiency.
  • The results show that tree-based models excel in certifiability, while gradient-based methods offer improved computational efficiency.

Read Full Article

like

1 Like

source image

Arxiv

2h

read

118

img
dot

Image Credit: Arxiv

CFIRE: A General Method for Combining Local Explanations

  • Researchers propose a new eXplainable AI algorithm for computing global decision rules.
  • The algorithm combines XAI methods with closed frequent itemset mining.
  • By addressing the disagreement problem, the algorithm accommodates different local explainers.
  • Evaluation of the algorithm demonstrates its robustness and improved performance in generating compact and complete rules.

Read Full Article

like

7 Likes

source image

Arxiv

2h

read

0

img
dot

Image Credit: Arxiv

Personalized Federated Training of Diffusion Models with Privacy Guarantees

  • The scarcity of accessible, compliant, and ethically sourced data presents a challenge for adopting AI in sensitive fields like healthcare and finance.
  • Diffusion models offer a solution for generating diverse synthetic data, which can be used as an alternative to restricted public datasets.
  • A novel federated learning framework is introduced for training diffusion models on decentralized private datasets.
  • The framework ensures robust differential privacy guarantees and produces high-quality samples, reducing biases and imbalances in synthetic data.

Read Full Article

like

Like

source image

Arxiv

2h

read

255

img
dot

Image Credit: Arxiv

MedPix 2.0: A Comprehensive Multimodal Biomedical Data set for Advanced AI Applications with Retrieval Augmented Generation and Knowledge Graphs

  • The paper introduces the MedPix 2.0 data set, a comprehensive multimodal biomedical data set for advanced AI applications.
  • The lack of high-quality medical data sets and the need for multimodal data sets due to the rise of large multimodal models are addressed.
  • The paper details the process of building the MedPix 2.0 data set, which involves extracting visual and textual data from the original MedPix data set.
  • In addition to the data set, a GUI and a knowledge graph extension are developed for efficient navigation and medical decision support.

Read Full Article

like

15 Likes

source image

Arxiv

2h

read

9

img
dot

Image Credit: Arxiv

Deep Learning-Based Hypoglycemia Classification Across Multiple Prediction Horizons

  • Type 1 diabetes (T1D) management can be enhanced through predictive machine learning algorithms.
  • This study integrates short- and long-term prediction horizons within a single classification model to enhance decision support.
  • LSTM models outperformed ResNet models in classifying nine classes, but longer prediction horizons remain challenging.
  • A population-based six-class model detected at least 60% of hypoglycemia events.

Read Full Article

like

Like

source image

Arxiv

2h

read

88

img
dot

Image Credit: Arxiv

I'm Sorry Dave: How the old world of personnel security can inform the new world of AI insider risk

  • Organisations are rapidly adopting artificial intelligence (AI) tools to perform tasks previously undertaken by people.
  • There is no meaningful interplay between the rapidly evolving domain of AI and the traditional world of personnel security, posing a problem.
  • Some concepts and approaches used to address security risks from human insiders are applicable to the emerging risks from AI insiders.
  • AI can be used defensively to protect against both human and AI insiders.

Read Full Article

like

5 Likes

source image

Arxiv

2h

read

104

img
dot

Image Credit: Arxiv

Enhance Vision-based Tactile Sensors via Dynamic Illumination and Image Fusion

  • Vision-based tactile sensors use structured light to measure deformation in their elastomeric interface.
  • This work explores the use of dynamic illumination patterns and image fusion techniques to enhance the quality of vision-based tactile sensors.
  • By capturing multiple measurements with different illumination patterns and fusing them together, a higher-quality measurement can be obtained.
  • Experimental results demonstrate significant improvements in image contrast, sharpness, and background difference, which can be retroactively applied to existing sensors or integrated into new hardware designs.

Read Full Article

like

6 Likes

source image

Arxiv

2h

read

108

img
dot

Image Credit: Arxiv

SandboxEval: Towards Securing Test Environment for Untrusted Code

  • Large language models (LLMs) have the potential to produce malicious code, posing risks to assessment infrastructure.
  • SandboxEval is a test suite designed to evaluate the security and confidentiality of test environments for LLM-generated code.
  • The suite focuses on vulnerabilities related to sensitive information exposure, filesystem manipulation, external communication, and other dangerous operations.
  • By deploying SandboxEval, developers can gain valuable insights to improve assessment infrastructure and identify risks associated with LLM execution.

Read Full Article

like

6 Likes

source image

Arxiv

2h

read

183

img
dot

Image Credit: Arxiv

Celler:A Genomic Language Model for Long-Tailed Single-Cell Annotation

  • Recent breakthroughs in single-cell technology have led to the need for efficient annotation of long-tailed single-cell data pertaining to disease conditions.
  • To address this challenge, Celler, a generative pre-training model, has been introduced that incorporates the Gaussian Inflation (GInf) Loss function and Hard Data Mining (HDM) strategy.
  • The GInf Loss function dynamically adjusts sample weights, improving the model's ability to learn from rare categories and reducing the risk of overfitting for common categories.
  • The HDM strategy targets difficult-to-learn minority data samples, significantly improving the model's predictive accuracy.

Read Full Article

like

11 Likes

source image

Arxiv

2h

read

196

img
dot

Image Credit: Arxiv

A multi-locus predictiveness curve and its summary assessment for genetic risk prediction

  • With the advancement of high-throughput genotyping and sequencing technologies, a need arises to evaluate the role of genetic predictors in disease prediction.
  • A multi-marker predictiveness curve is proposed to measure the combined effects of multiple genetic variants in risk prediction models for complex diseases.
  • The predictiveness curve is connected with the ROC curve and Lorenz curve.
  • The predictiveness U is introduced as a summary index to evaluate the predictive ability of risk prediction models, and it outperformed other summary indices in terms of unbiasedness and robustness.

Read Full Article

like

11 Likes

source image

Arxiv

2h

read

304

img
dot

Image Credit: Arxiv

Improving Diseases Predictions Utilizing External Bio-Banks

  • Machine learning can enhance disease predictions and uncover biologically meaningful associations, even with limited data.
  • LightGBM models trained on a dataset of 10K are used to impute metabolomics features.
  • Survival analysis is applied to assess the impact of imputed metabolomics features on disease-related risk factors.
  • Integration of survival analysis and genetic studies with machine learning can uncover valuable biomedical insights.

Read Full Article

like

18 Likes

source image

Arxiv

2h

read

62

img
dot

Image Credit: Arxiv

Imbalanced malware classification: an approach based on dynamic classifier selection

  • This study addresses the issue of class imbalance in malware detection on mobile devices.
  • The study evaluates various machine learning strategies for detecting malware in Android applications.
  • The proposed approach focuses on dynamic classifier selection algorithms, which have shown superior performance.
  • The empirical analysis demonstrates the effectiveness of the KNOP algorithm using a pool of Random Forest.

Read Full Article

like

3 Likes

source image

Arxiv

2h

read

186

img
dot

Image Credit: Arxiv

GAL-MAD: Towards Explainable Anomaly Detection in Microservice Applications Using Graph Attention Networks

  • The transition to microservices has revolutionized software architectures, offering enhanced scalability and modularity.
  • Anomaly detection is crucial for maintaining performance and functionality in microservice applications.
  • A novel anomaly detection model called GAL-MAD is proposed, leveraging Graph Attention and LSTM architectures.
  • GAL-MAD outperforms state-of-the-art models on the RS-Anomic dataset, achieving higher accuracy and recall.

Read Full Article

like

11 Likes

source image

Arxiv

2h

read

81

img
dot

Image Credit: Arxiv

Improving Predictions of Convective Storm Wind Gusts through Statistical Post-Processing of Neural Weather Models

  • Issuing timely severe weather warnings helps mitigate potentially disastrous consequences.
  • Neural Weather Models (NWMs) offer a computationally inexpensive and fast approach for forecasting atmospheric environments.
  • The study applied statistical and deep learning post-processing methods to forecast wind gusts using NWMs.
  • Results confirmed the added value of NWMs for extreme wind forecasting and designing more responsive early-warning systems.

Read Full Article

like

4 Likes

source image

Arxiv

2h

read

163

img
dot

Image Credit: Arxiv

Contextualize-then-Aggregate: Circuits for In-Context Learning in Gemma-2 2B

  • In-Context Learning (ICL) is an intriguing ability of large language models (LLMs).
  • Research finds that Gemma-2 2B uses a two-step strategy, contextualize-then-aggregate, for task information assembly.
  • In the lower layers, the model builds up representations of individual fewshot examples, contextualized by preceding examples.
  • In the higher layers, these representations are aggregated to identify the task and prepare predictions.

Read Full Article

like

9 Likes

For uninterrupted reading, download the app