Search:
Match:
23 results
research#unlearning📝 BlogAnalyzed: Jan 5, 2026 09:10

EraseFlow: GFlowNet-Driven Concept Unlearning in Stable Diffusion

Published:Dec 31, 2025 09:06
1 min read
Zenn SD

Analysis

This article reviews the EraseFlow paper, focusing on concept unlearning in Stable Diffusion using GFlowNets. The approach aims to provide a more controlled and efficient method for removing specific concepts from generative models, addressing a growing need for responsible AI development. The mention of NSFW content highlights the ethical considerations involved in concept unlearning.
Reference

“画像生成モデルもだいぶ進化を成し遂げており, それに伴って概念消去（unlearningに仮に分類しておきます）の研究も段々広く行われるようになってきました.”

PermalinkZenn SD
Research Paper#Federated Learning, Privacy, Unlearning🔬 ResearchAnalyzed: Jan 3, 2026 16:12

Certifying Data Removal in Federated Learning

Published:Dec 29, 2025 03:25
1 min read
ArXiv

Analysis

This paper addresses the critical issue of data privacy and the 'right to be forgotten' in vertical federated learning (VFL). It proposes a novel algorithm, FedORA, to efficiently and effectively remove the influence of specific data points or labels from trained models in a distributed setting. The focus on VFL, where data is distributed across different parties, makes this research particularly relevant and challenging. The use of a primal-dual framework, a new unlearning loss function, and adaptive step sizes are key contributions. The theoretical guarantees and experimental validation further strengthen the paper's impact.
Reference

“FedORA formulates the removal of certain samples or labels as a constrained optimization problem solved using a primal-dual framework.”

PermalinkArXiv
Research#llm🔬 ResearchAnalyzed: Dec 25, 2025 10:13

Investigating Model Editing for Unlearning in Large Language Models

Published:Dec 25, 2025 05:00
1 min read
ArXiv NLP

Analysis

This paper explores the application of model editing techniques, typically used for modifying model behavior, to the problem of machine unlearning in large language models. It investigates the effectiveness of existing editing algorithms like ROME, IKE, and WISE in removing unwanted information from LLMs without significantly impacting their overall performance. The research highlights that model editing can surpass baseline unlearning methods in certain scenarios, but also acknowledges the challenge of precisely defining the scope of what needs to be unlearned without causing unintended damage to the model's knowledge base. The study contributes to the growing field of machine unlearning by offering a novel approach using model editing techniques.
Reference

“model editing approaches can exceed baseline unlearning methods in terms of quality of forgetting depending on the setting.”

PermalinkArXiv NLP
Research#LLM🔬 ResearchAnalyzed: Jan 10, 2026 07:54

Model Editing for Unlearning: A Deep Dive into LLM Forgetting

Published:Dec 23, 2025 21:41
1 min read
ArXiv

Analysis

This research explores a critical aspect of responsible AI: how to effectively remove unwanted knowledge from large language models. The article likely investigates methods for editing model parameters to 'unlearn' specific information, a crucial area for data privacy and ethical considerations.
Reference

“The research focuses on investigating model editing techniques to facilitate 'unlearning' within large language models.”

PermalinkArXiv
Research#Unlearning🔬 ResearchAnalyzed: Jan 10, 2026 08:40

Machine Unlearning Explored in Quantum Machine Learning Context

Published:Dec 22, 2025 10:40
1 min read
ArXiv

Analysis

This ArXiv paper investigates the intersection of machine unlearning techniques and the emerging field of quantum machine learning. The empirical study likely assesses the effectiveness and challenges of removing specific data from quantum machine learning models.
Reference

“The paper is an empirical study.”

PermalinkArXiv
Research#llm🔬 ResearchAnalyzed: Jan 4, 2026 07:15

Feature-Selective Representation Misdirection for Machine Unlearning

Published:Dec 18, 2025 08:31
1 min read
ArXiv

Analysis

This article, sourced from ArXiv, likely presents a novel approach to machine unlearning. The title suggests a focus on selectively removing or altering specific features within a model's representation to achieve unlearning, which is a crucial area for privacy and data management in AI. The term "misdirection" implies a strategy to manipulate the model's internal representations to forget specific information.
Reference

“”

PermalinkArXiv
Research#llm🔬 ResearchAnalyzed: Jan 4, 2026 07:00

Dual-View Inference Attack: Machine Unlearning Amplifies Privacy Exposure

Published:Dec 18, 2025 03:24
1 min read
ArXiv

Analysis

This article discusses a research paper on a novel attack that exploits machine unlearning to amplify privacy risks. The core idea is that by observing the changes in a model after unlearning, an attacker can infer sensitive information about the data that was removed. This highlights a critical vulnerability in machine learning systems where attempts to protect privacy (through unlearning) can inadvertently create new attack vectors. The research likely explores the mechanisms of this 'dual-view' attack, its effectiveness, and potential countermeasures.
Reference

“The article likely details the methodology of the dual-view inference attack, including how the attacker observes the model's behavior before and after unlearning to extract information about the forgotten data.”

PermalinkArXiv
Research#llm🔬 ResearchAnalyzed: Jan 4, 2026 10:46

Erasing CLIP Memories: Non-Destructive, Data-Free Zero-Shot class Unlearning in CLIP Models

Published:Dec 16, 2025 06:37
1 min read
ArXiv

Analysis

This article likely presents a novel method for removing specific class information from CLIP models without requiring access to the original training data. The terms "non-destructive" and "data-free" suggest an efficient and potentially privacy-preserving approach to model updates. The focus on zero-shot unlearning indicates the method's ability to remove knowledge of classes not explicitly seen during the unlearning process, which is a significant advancement.
Reference

“The abstract or introduction of the ArXiv paper would provide the most relevant quote, but without access to the paper, a specific quote cannot be provided. The core concept revolves around removing class-specific knowledge from a CLIP model without retraining or using the original training data.”

PermalinkArXiv
Research#CLIP🔬 ResearchAnalyzed: Jan 10, 2026 10:52

Unlearning for CLIP Models: A Novel Training- and Data-Free Approach

Published:Dec 16, 2025 05:54
1 min read
ArXiv

Analysis

This research explores a novel method for unlearning in CLIP models, crucial for addressing data privacy and model bias. The data-free approach could significantly enhance the flexibility and applicability of these models across various domains.
Reference

“The research focuses on selective, controlled, and domain-agnostic unlearning.”

PermalinkArXiv
Research#llm🔬 ResearchAnalyzed: Jan 4, 2026 09:51

Dual-Phase Federated Deep Unlearning via Weight-Aware Rollback and Reconstruction

Published:Dec 15, 2025 14:32
1 min read
ArXiv

Analysis

This article, sourced from ArXiv, likely presents a novel approach to federated deep unlearning. The title suggests a two-phase process that leverages weight-aware rollback and reconstruction techniques. The focus is on enabling models to 'forget' specific data in a federated learning setting, which is crucial for privacy and compliance. The use of 'weight-aware' implies a sophisticated method that considers the importance of different weights during the unlearning process. The paper's contribution would be in improving the efficiency, accuracy, or privacy guarantees of unlearning in federated learning.
Reference

“The paper likely addresses the challenge of removing the influence of specific data points from a model trained in a federated setting, while preserving the model's performance on the remaining data.”

PermalinkArXiv
Research#LLM🔬 ResearchAnalyzed: Jan 10, 2026 11:08

FROC: A Novel Framework for Machine Unlearning in Large Language Models

Published:Dec 15, 2025 13:53
1 min read
ArXiv

Analysis

The paper introduces FROC, a framework aimed at improving machine unlearning capabilities in Large Language Models. This is a critical area for responsible AI development, focusing on data removal and model adaptation.
Reference

“FROC is a unified framework with risk-optimized control.”

PermalinkArXiv
Research#Face Retrieval🔬 ResearchAnalyzed: Jan 10, 2026 11:09

Unlearning Face Identity for Enhanced Retrieval Systems

Published:Dec 15, 2025 13:35
1 min read
ArXiv

Analysis

This research explores a novel method for improving retrieval systems by removing face identity information. The approach, detailed in an ArXiv paper, likely focuses on privacy-preserving techniques while potentially boosting efficiency.
Reference

“The research is based on a paper from ArXiv.”

PermalinkArXiv
Research#MLLM🔬 ResearchAnalyzed: Jan 10, 2026 11:48

Machine Unlearning for Multimodal Large Language Models using Visual Knowledge Distillation

Published:Dec 12, 2025 06:51
1 min read
ArXiv

Analysis

This research explores a crucial area: enabling multimodal LLMs to forget specific information, which is essential for data privacy and model adaptability. The method, using visual knowledge distillation, provides a promising approach to address the challenge of machine unlearning in complex models.
Reference

“The research focuses on machine unlearning for multimodal LLMs.”

PermalinkArXiv
Research#Federated Learning🔬 ResearchAnalyzed: Jan 10, 2026 12:06

REMISVFU: Federated Unlearning with Representation Misdirection

Published:Dec 11, 2025 07:05
1 min read
ArXiv

Analysis

This research explores federated unlearning in a vertical setting using a novel representation misdirection technique. The core concept likely focuses on how to remove or mitigate the impact of specific data points from a federated model while preserving its overall performance.
Reference

“The article's context indicates the research is published on ArXiv, suggesting a focus on academic novelty.”

PermalinkArXiv
Research#Unlearning🔬 ResearchAnalyzed: Jan 10, 2026 12:15

MedForget: Advancing Medical AI Reliability Through Unlearning

Published:Dec 10, 2025 17:55
1 min read
ArXiv

Analysis

This ArXiv paper introduces a significant contribution to the field of medical AI by proposing a hierarchy-aware multimodal unlearning testbed. The focus on unlearning, crucial for data privacy and model robustness, is highly relevant given growing concerns around AI in healthcare.
Reference

“The paper focuses on a 'hierarchy-aware multimodal unlearning testbed'.”

PermalinkArXiv
Research#LLM🔬 ResearchAnalyzed: Jan 10, 2026 12:48

LUNE: Fast and Effective LLM Unlearning with Negative Examples

Published:Dec 8, 2025 10:10
1 min read
ArXiv

Analysis

This research explores efficient methods for 'unlearning' information from Large Language Models, which is crucial for data privacy and model updates. The use of LoRA fine-tuning with negative examples provides a novel approach to achieving this, potentially accelerating the model's ability to forget unwanted data.
Reference

“The research utilizes LoRA fine-tuning with negative examples to achieve efficient unlearning.”

PermalinkArXiv
Research#LLM🔬 ResearchAnalyzed: Jan 10, 2026 12:48

Efficient LLM Unlearning: Gradient Reconstruction from LoRA for Privacy

Published:Dec 8, 2025 10:10
1 min read
ArXiv

Analysis

This research explores a novel method for efficiently unlearning information from Large Language Models (LLMs) using gradient reconstruction from LoRA. The approach offers potential for improving model privacy and compliance with data removal requests.
Reference

“Gradient Reconstruction from LoRA”

PermalinkArXiv
Research#LLM🔬 ResearchAnalyzed: Jan 10, 2026 13:15

RapidUn: Efficient Unlearning for Large Language Models via Parameter Reweighting

Published:Dec 4, 2025 05:00
1 min read
ArXiv

Analysis

The research paper explores a method for efficiently unlearning information from large language models, a critical aspect of model management and responsible AI. Focusing on parameter reweighting offers a potentially faster and more resource-efficient approach compared to retraining or other unlearning strategies.
Reference

“The paper focuses on influence-driven parameter reweighting for efficient unlearning.”

PermalinkArXiv
Research#Bias🔬 ResearchAnalyzed: Jan 10, 2026 13:42

Debiasing Sonar Image Classification: A Supervised Contrastive Unlearning Approach

Published:Dec 1, 2025 05:25
1 min read
ArXiv

Analysis

This research explores a crucial problem in AI: mitigating bias in image classification, specifically within a specialized domain (sonar). The supervised contrastive unlearning technique and explainable AI aspects suggest a focus on both accuracy and transparency, which is valuable for practical application.
Reference

“The research focuses on the problem of background bias in sonar image classification.”

PermalinkArXiv
Research#LLMs🔬 ResearchAnalyzed: Jan 10, 2026 14:14

Reasoning-Preserving Unlearning in Multimodal LLMs Explored

Published:Nov 26, 2025 13:45
1 min read
ArXiv

Analysis

This ArXiv article likely investigates methods for removing information from multimodal large language models while preserving their reasoning abilities. The research addresses a crucial challenge in AI, ensuring models can be updated and corrected without losing core functionality.
Reference

“The context indicates an ArXiv article exploring unlearning in multimodal large language models.”

PermalinkArXiv
Research#llm🔬 ResearchAnalyzed: Jan 4, 2026 09:23

Geometric-Disentangelment Unlearning

Published:Nov 21, 2025 09:58
1 min read
ArXiv

Analysis

This article likely discusses a novel approach to unlearning in machine learning, specifically focusing on geometric and disentanglement aspects. The title suggests a method to remove or mitigate the influence of specific data points or concepts from a model by manipulating its geometric representation and disentangling learned features. The use of "unlearning" implies a focus on privacy, data deletion, or model adaptation.

Key Takeaways

    Reference

    “”

    PermalinkArXiv
    Research#llm🔬 ResearchAnalyzed: Jan 4, 2026 08:59

    Forgetting-MarI: LLM Unlearning via Marginal Information Regularization

    Published:Nov 14, 2025 22:48
    1 min read
    ArXiv

    Analysis

    This article introduces a method called Forgetting-MarI for LLM unlearning. The core idea is to use marginal information regularization to help LLMs forget specific information. The paper likely explores the effectiveness and efficiency of this approach compared to other unlearning techniques. The focus is on improving the privacy and adaptability of LLMs.
    Reference

    “”

    PermalinkArXiv
    Research#AI/Machine Learning📝 BlogAnalyzed: Jan 3, 2026 06:13

    Concept Erasure from Stable Diffusion: CURE (Paper)

    Published:Oct 19, 2025 09:34
    1 min read
    Zenn SD

    Analysis

    The article announces a paper accepted at NeurIPS 2025, focusing on concept unlearning in diffusion models. It introduces the CURE method, referencing the paper by Biswas, Roy, and Roy. The article provides a brief overview, likely setting the stage for a deeper dive into the research.
    Reference

    “CURE: Concept unlearning via orthogonal representation editing in Diffusion Models (NeurIPS2025) and the paper by Shristi Das Biswas, Arani Roy, and Kaushik Roy.”

    PermalinkZenn SD