Charles Casimiro Cavalcante

Research profile

Research summary

Charles Casimiro Cavalcante is a Brazilian researcher at Universidade Federal do Ceará whose record spans roughly a quarter century of signal processing and machine learning research. The bibliometric snapshot, resolved from OpenAlex (ORCID 0000-0002-4198-4064), shows a continuous by-year output series from 2000 to 2026, peaking at 12 items in 2013, with 21 works in the most recent three years and an activity score of 90 — the strongest signal in this profile. Topic counts place advanced wireless communication techniques (51), advanced MIMO systems optimization (37), wireless communication networks (28), adaptive filtering (23), and blind source separation (21) at the center of the record, while the recent field distribution splits evenly between optimal-transport-based unsupervised clustering and federated learning. Two 2026 arXiv preprints reviewed by Evolve list him as last author: a multi-domain clustering method that learns shared prototypes by minimizing Sinkhorn and maximum mean discrepancy divergence, and pFedKDH, a personalized federated learning scheme combining backbone-only sharing with recalibrated-head distillation. Both are rated medium for reproducibility (mean 55), with noted gaps: no seeds, significance tests, or code in the former, and possible inflation from server-side labeled data in the latter. The snapshot lists an h-index of 16 and an i10-index of 21; consolidated output and impact dimension scores remain uncomputed.

Research keywords

Highlights

• Shared cluster prototypes via Sinkhorn and MMD divergence across domains

  In Multi-Domain Clustering via Measure Quantization (2026, arXiv), this last-authored paper learns cluster prototypes shared across domains by minimizing Sinkhorn and maximum mean discrepancy divergence, reporting strong benchmark wins; Evolve notes the absence of seeds, significance tests, and code.

• pFedKDH: personalized federated learning with distillation and local heads

  The 2026 preprint shares only the backbone globally while distilling into recalibrated local classifier heads, winning most label-skew federated learning benchmarks; Evolve cautions that server-side labeled data may inflate the reported gains.

• Two-decade wireless signal processing research base

  Topic counts in the snapshot are led by advanced wireless communication techniques (51), advanced MIMO systems optimization (37), and wireless communication networks (28), with adaptive filtering (23) and blind source separation (21) — the statistical signal processing base beneath the recent machine learning work.

Timeline

1. 2000Earliest year in the provided by-year bibliometric series (2 recorded items); the record continues unbroken through 2026.
2. 2013The by-year series reaches its maximum of 12 recorded items, the highest annual value in the 2000–2026 span.
3. 20248 recorded items, opening a strong recent stretch that totals 21 works across the last three years.
4. 2026Two last-authored arXiv preprints enter the Evolve catalog: multi-domain clustering via measure quantization and the pFedKDH personalized federated learning method.

Papers in the Evolve catalog

• 2026-09-18

  Multi-Domain Clustering via Measure Quantization

• 2026-09-15

  Personalized Federated Learning through Global Knowledge Distillation and Local Head Adaptation

Frequent co-authors