Attention to Lightning Change: Utilizing BERTs to Detect Bitcoin Change Address with Lightning Network Information
Bibliographic record
Abstract
As the popularity of Bitcoin has increased over the years, the limited capacity of its blocks has led to competitive transaction fee bidding among users. This structural limitation undermines Bitcoin’s suitability for microtransactions due to increasing transaction fees. In response, Poon and Dryja introduced the Lightning Network (LN), a second-layer solution that provides almost instant transactions with minimal fees. Over the past few years, LN has attracted more and more users, currently securing over $336 million in Bitcoin within its channels to date. However, the expansion of the Lightning Network brings potential challenges. Among them is the risk to compromise user anonymity by creating public LN channels. While existing studies primarily explore privacy within the LN or Bitcoin network independently, cross-layer privacy remains underexplored. In this paper, we present concrete evidence to showcase the potential effect of using the Lightning Network on the anonymity of Bitcoin users. Our proposed solution demonstrates how data collected from the Layer 2 Lightning Network can be used to detect change addresses residing in the Layer 1 Bitcoin network. First, we analyze the distribution of transaction inputs and outputs and compare transactions in a range of Bitcoin blocks, with those linked to LN channel creation. We then implement a machine learning model utilizing an encoder-only transformer to identify change outputs in Bitcoin transactions. Lastly, we compared the results to show the potential of cross-chain data. Our model successfully achieved its goal by flagging the correct change output with 94% accuracy with the cross-chain data, compared to 87% baseline which is a significant improvement. Our result provides valuable insights into the impact of the Lightning Network on Bitcoin user privacy and sets the groundwork for further investigation into cross-layer deanonymization.
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.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.008 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.002 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
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 itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".