Backdoored model code enables deterministic, verifiable stealing of sparse secrets during local LLM fine-tuning via tensor-rule matching and gradient injection, achieving over 98% strict attack success rate while bypassing DP-SGD and auditing defenses.
hub
write newline
15 Pith papers cite this work. Polarity classification is still indexing.
15
Pith papers citing it
hub tools
verdicts
UNVERDICTED 15representative citing papers
A method infers influential instances for data cleansing in SGD-trained models by retracing training steps with intermediate models, shown to improve performance on MNIST and CIFAR10.
BITRec improves generative multi-behavior recommendation by modeling behavioral intensity via separated pathways and transitions via learnable relation matrices, reporting 15-23% gains on large retail datasets.
MPSF-FL combines multi-problem pre-training with federated learning to improve performance and generalization of neural solvers across diverse vehicle routing problems.
ArgLLMs build argumentation frameworks from LLMs to support explainable and contestable formal reasoning for claim verification.
GACG infers a coordination graph capturing both pair and group dependencies for information exchange in MARL, adds a group distance loss for consistency, and reports superior performance on StarCraft II micromanagement tasks.
ECDF-based anytime performance metrics improve both analysis of MaxSAT local search solver behavior and automatic hyperparameter configuration compared to final-fitness metrics.
Bayesian uncertainty matching aligns joint feature-label distributions to improve unsupervised domain adaptation and reduce negative transfer on benchmark datasets.
Ister is a linear-complexity transformer using Dot-attention and inverted seasonal-trend decomposition for multivariate time series forecasting that reports state-of-the-art benchmark performance.
MRMN is a unified neural framework using collaborative metric learning and memory networks to model fine-grained relations from multiple user feedback types and outperform prior recommender systems.
Formalizes collective mobile sequential recommendation for multiple taxicabs, defines a new metric minimizing sum of potential travel time, and shows a greedy algorithm outperforming conventional methods in trace-driven simulations.
A greedy submodular maximization method for mini-batch selection in DNN training yields better generalization than SGD on standard datasets.
A literature survey that introduces a taxonomy for computational and communication efficiency in federated learning with foundation models and discusses PEFT, framework readiness, and open research questions.
OPPO augments PPO with optimistic policy evaluation driven by return uncertainty estimates and shows improved results over prior methods on a tabular sparse-reward task.
Authors introduce a Conflict model emphasizing sequence repulsion and report empirical gains when combined with attention on NLU tasks.
citing papers explorer
-
Secret Stealing Attacks on Local LLM Fine-Tuning through Supply-Chain Model Code Backdoors
Backdoored model code enables deterministic, verifiable stealing of sparse secrets during local LLM fine-tuning via tensor-rule matching and gradient injection, achieving over 98% strict attack success rate while bypassing DP-SGD and auditing defenses.
-
Data Cleansing for Models Trained with SGD
A method infers influential instances for data cleansing in SGD-trained models by retracing training steps with intermediate models, shown to improve performance on MNIST and CIFAR10.
-
Modeling Behavioral Intensity and Transitions for Generative Recommendation
BITRec improves generative multi-behavior recommendation by modeling behavioral intensity via separated pathways and transitions via learnable relation matrices, reporting 15-23% gains on large retail datasets.
-
Enhancing Cross-Problem Vehicle Routing via Federated Learning
MPSF-FL combines multi-problem pre-training with federated learning to improve performance and generalization of neural solvers across diverse vehicle routing problems.
-
Argumentative Large Language Models for Explainable and Contestable Claim Verification
ArgLLMs build argumentation frameworks from LLMs to support explainable and contestable formal reasoning for claim verification.
-
Group-Aware Coordination Graph for Multi-Agent Reinforcement Learning
GACG infers a coordination graph capturing both pair and group dependencies for information exchange in MARL, adds a group distance loss for consistency, and reports superior performance on StarCraft II micromanagement tasks.
-
Better Understandings and Configurations in MaxSAT Local Search Solvers via Anytime Performance Analysis
ECDF-based anytime performance metrics improve both analysis of MaxSAT local search solver behavior and automatic hyperparameter configuration compared to final-fitness metrics.
-
Bayesian Uncertainty Matching for Unsupervised Domain Adaptation
Bayesian uncertainty matching aligns joint feature-label distributions to improve unsupervised domain adaptation and reduce negative transfer on benchmark datasets.
-
Ister: Linear Transformer for Efficient Multivariate Time Series Forecasting
Ister is a linear-complexity transformer using Dot-attention and inverted seasonal-trend decomposition for multivariate time series forecasting that reports state-of-the-art benchmark performance.
-
Collaborative Metric Learning with Memory Network for Multi-Relational Recommender Systems
MRMN is a unified neural framework using collaborative metric learning and memory networks to model fine-grained relations from multiple user feedback types and outperform prior recommender systems.
-
Collective Mobile Sequential Recommendation: A Recommender System for Multiple Taxicabs
Formalizes collective mobile sequential recommendation for multiple taxicabs, defines a new metric minimizing sum of potential travel time, and shows a greedy algorithm outperforming conventional methods in trace-driven simulations.
-
Submodular Batch Selection for Training Deep Neural Networks
A greedy submodular maximization method for mini-batch selection in DNN training yields better generalization than SGD on standard datasets.
-
A Survey on Efficient Federated Learning Methods for Foundation Model Training
A literature survey that introduces a taxonomy for computational and communication efficiency in federated learning with foundation models and discusses PEFT, framework readiness, and open research questions.
-
Optimistic Proximal Policy Optimization
OPPO augments PPO with optimistic policy evaluation driven by return uncertainty estimates and shows improved results over prior methods on a tabular sparse-reward task.
-
Conflict as an Inverse of Attention in Sequence Relationship
Authors introduce a Conflict model emphasizing sequence repulsion and report empirical gains when combined with attention on NLU tasks.