FedITD: A Federated Parameter-Efficient Tuning With Pre-Trained Large Language Models and Transfer Learning Framework for Insider Threat Detection
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Insider threats cause greater losses than external attacks, prompting organizations to invest in detection systems. However, there exist challenges: 1) Security and privacy concerns prevent data sharing, making it difficult to train robust models and identify new attacks. 2) The diversity and uniqueness of organizations require localized models, as a universal solution could be more effective. 3) High resource costs, delays, and data security concerns complicate building effective detection systems. This paper introduces FedITD, a flexible, hierarchy, and federated framework with local real-time detection systems, combining Large Language Models (LLM), Federated Learning (FL), Parameter Efficient Tuning (PETuning), and Transfer Learning (TF) for insider threat detection. FedITD uses FL to protect privacy while indirect integrating client information and employs PETuning methods (Adapter, BitFit, LoRA) with LLMs (BERT, RoBERTa, XLNet, DistilBERT) to reduce resource use and time delay. FedITD customizes client models and optimizes performance via transfer learning without central data transfer, further enhancing the detection of new attacks. FedITD outperforms other federated learning methods and its performance is very close to the best centrally trained method. Extensive experiment results show FedITD’s superior performance, adaptability to varied data, and reduction of resource costs, achieving an optimal balance in detection capabilities across source data, unlabeled local data, and global data. Alternative PETuning implementations are also explored in this paper.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.002 | 0.001 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it