Lester James V. Miranda
Scholar

Lester James V. Miranda

Google Scholar ID: 2RtnNKEAAAAJ
University of Cambridge
Natural Language ProcessingMachine Learning
Citations & Impact
All-time
Citations
1,736
 
H-index
11
 
i10-index
12
 
Publications
20
 
Co-authors
9
list available
Resume (English only)
Academic Achievements
  • FilBench: A comprehensive LLM benchmark for Filipino, accepted at EMNLP 2025 Main
  • Three first/co-first author papers accepted at ACL 2025 Main: HyPER, M-RewardBench, and UD-NewsCrawl
  • Collaborative project SEA-VL also accepted at ACL 2025 Main
  • Contributed to Tülu 3 and OLMo 2 releases by scaling preference data via a synthetic on-policy pipeline, improving DPO models
  • Co-led (with Yizhong Wang) the paper 'Hybrid Preferences' on routing preference instances to human or LM annotators
  • Collaborated on 'M-RewardBench', evaluating reward models in multilingual settings, with researchers from Cohere for AI