Yangsibo Huang, Kai Li, Danqi Chen, Samyak Gupta, Zexuan Zhong, Tianyu Gao
We lifted 8 functions out of this paper's own repositories and ran 5 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.
| Repository | Role | Ran |
|---|---|---|
| princeton-sysml/film | canonical | 5 of 8 |
| Function | Status | Where it lives |
|---|---|---|
| calculatePerplexity | Ran | princeton-sysml/film/reorder.py code served (permissive licence) · get_code("2506162f964eed68") |
| ibis | Ran | princeton-sysml/film/reorder.py code served (permissive licence) · get_code("30062ec641d7c6e1") |
| shard_sequence_using_eos | Ran | princeton-sysml/film/reorder.py code served (permissive licence) · get_code("09f1ec0ed6850234") |
| shuffle_proposals | Ran | princeton-sysml/film/reorder.py code served (permissive licence) · get_code("7b09c74c1f6987a0") |
| shuffle_proposals_rand | Ran | princeton-sysml/film/reorder.py code served (permissive licence) · get_code("6d0e59be9216d2ac") |
| get_grad_gpt2 | Not yet run | princeton-sysml/film/reorder.py code served (permissive licence) · get_code("7cecdfca7a210f88") |
| ibisAlgorithm | Not yet run | princeton-sysml/film/reorder.py code served (permissive licence) · get_code("45a5586b7fa0bff9") |
| score | Not yet run | princeton-sysml/film/reorder.py code served (permissive licence) · get_code("b902d2253e571934") |
Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.
Federated learning allows distributed users to collaboratively train a model while keeping each user's data private. Recently, a growing body of work has demonstrated that an eavesdropping attacker can effectively recover image data from gradients transmitted during federated learning. However, little progress has been made in recovering text data. In this paper, we present a novel attack method FILM for federated learning of language models (LMs). For the first time, we show the feasibility of recovering text from large batch sizes of up to 128 sentences. Unlike image-recovery methods that are optimized to match gradients, we take a distinct approach that first identifies a set of words from gradients and then directly reconstructs sentences based on beam search and a prior-based reordering strategy. We conduct the FILM attack on several large-scale datasets and show that it can successfully reconstruct single sentences with high fidelity for large batch sizes and even multiple sentences if applied iteratively. We evaluate three defense methods: gradient pruning, DPSGD, and a simple approach to freeze word embeddings that we propose. We show that both gradient pruning and DPSGD lead to a significant drop in utility. However, if we fine-tune a public pre-trained LM on private text without updating word embeddings, it can effectively defend the attack with minimal data utility loss. Together, we hope that our results can encourage the community to rethink the privacy concerns of LM training and its standard practices in the future. 2 * The first two authors contributed equally.
The same record, over MCP at https://syntology.ai/mcp:
get_harvested_code_for_paper("2205.08514")
get_code_for_paper("2205.08514")
have("2205.08514")
Connect an agent — have() is free.