{"paper":{"title":"State-of-art minibatches via novel DPP kernels: discretization, wavelets, and rough objectives","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Wavelet-based DPPs on Euclidean space discretize to low-rank kernels that preserve superior variance reduction for minibatches on rough objectives.","cross_cats":["cs.LG","math.PR"],"primary_cat":"stat.ML","authors_text":"Hoang-Son Tran, Pranav Gupta, R\\'emi Bardenet, Subhroshekhar Ghosh","submitted_at":"2026-05-13T07:54:37Z","abstract_excerpt":"Determinantal point processes (DPPs) have emerged as a kernelized alternative to vanilla independent sampling for generating efficient minibatches, coresets and other parsimonious representations of large-scale datasets. While theoretical foundations and promising empirical performance have been demonstrated, there are two challenges for current proposals for DPP-based coresets or minibatches. The first is the need for families of DPPs with certain key variance reduction properties, usually constructed in a continuous setting, of which there are few known examples. The second is the need for a"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We propose new DPPs on the Euclidean space based on wavelets, with provably better accuracy guarantees than the best known rates. Second, we introduce a general method to convert such continuous DPPs into discrete kernels, which simultaneously preserves the desired variance decay and reveals a low-rank decomposition of the discrete kernel.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The discretization procedure preserves the variance reduction properties of the continuous wavelet DPPs with only negligible degradation when applied to finite datasets, and that the low-rank structure remains exploitable without hidden computational costs.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Wavelet DPP kernels deliver improved continuous variance reduction and a discretization procedure that preserves decay rates for discrete ML subsampling tasks including rough objectives.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Wavelet-based DPPs on Euclidean space discretize to low-rank kernels that preserve superior variance reduction for minibatches on rough objectives.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"b3b6cfae40857457276f3da9d3d30e7db947391561ce3c9c0fc6006054397c30"},"source":{"id":"2605.13127","kind":"arxiv","version":1},"verdict":{"id":"b75d245f-ec28-4bf7-8395-8a48dee0a16e","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T18:23:15.698803Z","strongest_claim":"We propose new DPPs on the Euclidean space based on wavelets, with provably better accuracy guarantees than the best known rates. Second, we introduce a general method to convert such continuous DPPs into discrete kernels, which simultaneously preserves the desired variance decay and reveals a low-rank decomposition of the discrete kernel.","one_line_summary":"Wavelet DPP kernels deliver improved continuous variance reduction and a discretization procedure that preserves decay rates for discrete ML subsampling tasks including rough objectives.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The discretization procedure preserves the variance reduction properties of the continuous wavelet DPPs with only negligible degradation when applied to finite datasets, and that the low-rank structure remains exploitable without hidden computational costs.","pith_extraction_headline":"Wavelet-based DPPs on Euclidean space discretize to low-rank kernels that preserve superior variance reduction for minibatches on rough objectives."},"references":{"count":299,"sample":[{"doi":"","year":null,"title":"Miramont, J. M. and Tan, K. A. and Mukherjee, S. S. and Bardenet, R. and Ghosh, S. , date-added =. arXiv preprint arXiv:2504.07720 , title =","work_id":"1b079da6-1530-4530-9466-ddb967f59414","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Miramont, J. M. and Auger, F. and Colominas, M. A. and Laurent, N. and Meignen, S. , date-added =. Signal Processing , title =","work_id":"919a03eb-7645-450a-a33a-8d692a141907","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Miramont, J. M. and Bardenet, R. and Chainais, P. and Auger, F. , booktitle =. Adaptive hyperparameter tuning for time-frequency algorithms based on the zeros of the spectrogram , year =","work_id":"249c1ea2-ebe6-440f-ad2f-843769014c62","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Lacoste--Julien, S. and Husz. Approximate inference for the loss-calibrated. Proceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics (AISTATS) , organization =","work_id":"ae3fb55e-6d4b-45f6-9c4b-ddfee0def6db","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Advances in Neural Information Processing Systems (NeurIPS) , title =","work_id":"bc34f26f-676a-459b-919b-5d9fba8529b3","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":299,"snapshot_sha256":"00d7e81a1190c82abae8ac2c0e16cdd82fe8752df2535c9aa15a3c3ce0b0e919","internal_anchors":16},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}