Research map / Learning theory and algorithms / Clustering
Discrete Continuous Optimization Framework for Simultaneous Clustering and Training in Mixture Models
Parth Vipul Sangani, Arjun Shashank Kashettiwar, Pritish Chakraborty, Bhuvan Reddy Gangula, Durga S, Ganesh Ramakrishnan, Rishabh K Iyer, Abir De
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Joint discrete-continuous partitioning and training (PRESTO) instead of MoE / unsupervised partitioning | mixture of ResNet18-reduced components · CIFAR10 | Accuracy: 95.49 vs 90.02 (MoE, second best) | Improved |
| Joint discrete-continuous partitioning and training (PRESTO) instead of baselines | mixture of ResNet18-reduced components · CIFAR100 | Accuracy: 94.86 vs 85.19 (MoE, second best) | Improved |
| Joint discrete-continuous partitioning and training (PRESTO) instead of baselines | mixture of ResNet18-reduced components · STL | Accuracy: 90.07 vs 81.93 (Learn-MLR, second best) | Improved |
| Joint partitioning/training (PRESTO) vs knowledge distillation and pruning in resource-constrained setting | PRESTO lightweight partitioned models · CIFAR10 | GPU memory during training (MiB): 58.12 vs 274.27 (KD) | Improved |
| Partitioned lightweight models (PRESTO) vs KD at inference | PRESTO lightweight partitioned models · CIFAR10 | GPU memory at test (MiB): 12.07 vs 27.61 (KD) | Improved |
| Increasing number of partitions K from 4 to 7 (ablation) | PRESTO · CIFAR10, CIFAR100, STL | Accuracy | Mixed |
| Accuracy-vs-K sweep for PRESTO (partition count) (ablation) | PRESTO · CIFAR10 | Accuracy | Mixed |
| Joint partitioning and training (PRESTO) vs Agglomerative clustering on synthetic data | PRESTO · Synthetic degenerate mixture | ARI: 0.92 vs 0.79 (Agglomerative) | Improved |
| Joint partitioning and training (PRESTO) vs Kmeans++ / Learn-MLR for partition quality on CIFAR10/CIFAR100 | PRESTO · CIFAR100 | ICD (inter-cluster dissimilarity): PRESTO 0.712 vs MoE 0.743 | Mixed |
Closest papers by meaning
- Data Mixture Optimization: A Multi-fidelity Multi-scale Bayesian Framework (NeurIPS 2025 · similarity 0.86)
- Data-Efficient Learning via Clustering-Based Sensitivity Sampling: Foundation Models and Beyond (ICML 2024 · similarity 0.84)
- MixMin: Finding Data Mixtures via Convex Minimization (ICML 2025 · similarity 0.84)
- Multinoulli Extension: A Lossless Yet Effective Probabilistic Framework for Subset Selection over Partition Constraints (ICML 2025 · similarity 0.84)
- Big Learning Expectation Maximization (AAAI 2024 · similarity 0.84)
- Scaling Laws for Optimal Data Mixtures (NeurIPS 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
