Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
Differentially Private Learning with Adaptive Clipping
Authors: Galen Andrew, Om Thakkar, Brendan McMahan, Swaroop Ramaswamy
NeurIPS 2021 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments demonstrate that adaptive clipping to the median update norm works well across a range of realistic federated learning tasks, sometimes outperforming even the best fixed clip chosen in hindsight, and without the need to tune any clipping hyperparameter. To empirically validate the approach, we examine the behavior of our algorithm on six of the public benchmark federated learning tasks defined by Reddi et al. [24] |
| Researcher Affiliation | Industry | Galen Andrew EMAIL Om Thakkar EMAIL H. Brendan Mc Mahan EMAIL Swaroop Ramaswamy EMAIL |
| Pseudocode | Yes | Algorithm 1 DPFed Avg-M with adaptive clipping |
| Open Source Code | Yes | The code used for all of our experiments is publicly available at https://github.com/googleresearch/federated/blob/master/differential_privacy/run_federated.py. |
| Open Datasets | Yes | on six of the public benchmark federated learning tasks defined by Reddi et al. [24] |
| Dataset Splits | Yes | For all configurations, we report the best performing model whose server learning rate was chosen from this small grid on the validation set.4 |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU or CPU models used for the experiments. |
| Software Dependencies | No | The paper does not provide specific software dependencies with version numbers. |
| Experiment Setup | Yes | Hyperparameters are as discussed in the text and used in the experiments of Section 3: ηC = 0.2, C0 = 0.1, m = 100, σb = m/20. The optimal baseline client and server learning rates for our experimental setup are shown in Table 1 (right). |