Notice: This website is continuously updated as new conference proceedings and journal volumes are published. Consequently, the figures presented here may differ from those reported in our publications.
A brief overview of our methodology is provided on the Methods page, along with references to publications documenting our approach and links to the code and data used in each study.
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.
Probably Approximately Correct Constrained Learning
Authors: Luiz Chamon, Alejandro Ribeiro
NeurIPS 2020 | Venue PDF | LLM Run Details | Input Tokens: 19,679 Total number of tokens sent to the LLM as input for this paper's analysis. | Output Tokens: 5,056 Total number of tokens produced by the LLM (including reasoning/thinking tokens) for this paper's analysis.
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We analyze the generalization properties of this solution and use it to illustrate how constrained learning can address problems in fair and robust classification. 6 Numerical experiments Due to space constraints, we only provide highlights of the results obtained for the problems from Section 3. For more details and additional experiments, see Appendix D in the extended version [56]. |
| Researcher Affiliation | Academia | Luiz F. O. Chamon Dept. of Electrical and Systems Engineering University of Pennsylvania Pennsylvania, USA EMAIL Alejandro Ribeiro Dept. of Electrical and Systems Engineering University of Pennsylvania Pennsylvania, USA EMAIL |
| Pseudocode | Yes | Algorithm 1 Primal-dual near-PACC learner |
| Open Source Code | No | No explicit statement about releasing the source code for the described methodology or a link to a code repository is provided in the paper. |
| Open Datasets | Yes | In the Adult dataset [70], our goal is to predict whether an individual makes more than US$ 50,000.00 while being insensitive to gender. In this illustration, we use Algorithm 1 to train a Res Net18 [73] to classify images from the FMNIST dataset [74]. |
| Dataset Splits | No | The best accuracy over the validation set is achieved after 67 epochs, yielding a solution with test accuracy of 93.5% (Figure 2a). |
| Hardware Specification | No | The paper does not specify the hardware used for running the experiments (e.g., specific CPU/GPU models, memory). |
| Software Dependencies | No | For step 3 of Algorithm 1, we use ADAM [71] with batch size 128 and learning rate 0.1. |
| Experiment Setup | Yes | For step 3 of Algorithm 1, we use ADAM [71] with batch size 128 and learning rate 0.1. All other parameters were kept as in the original paper. After each epoch, we update the dual variables (step 4), also using ADAM with a step size of 0.01. All classifiers were trained over 300 epochs. |