Samsung ML researcher: parameter-efficient fine-tuning and budget-aware LoRA merging for post-RL reasoning LLMs
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
5
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-16) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.
Research profile
Output—
Not yet scored. This axis would measure publication volume and cadence (career totals, works per year, authorship roles); catalog data is still insufficient to compute it, though sparse OpenAlex yearly counts of 1-5 works since 2020 are on file.
Impact—
Not yet scored. This axis would measure citation-based influence (h-index, i10-index, per-paper citations); current signals are thin - an h-index of 5 and i10-index of 3 - so no score is computed until coverage improves.
Focus0/100
Scores 0/100, reflecting a split signal rather than an empty record. Within the Evolve catalog the field distribution is evenly divided (0.5 LLM fine-tuning, 0.5 model merging/PEFT), but the OpenAlex topic profile spans five clusters - Topic Modeling (7), Information Retrieval and Search Behavior (6), Geothermal Energy Systems (6), Integrated Energy Systems Optimization (4), Natural Language Processing Techniques (3) - suggesting breadth or name-disambiguation noise; the reviewed papers themselves are tightly LoRA-centric.
Collaboration39/100
Scores 39/100 on a compact network: 6 distinct coauthors, matching the six recurring names - Lazar Valkov, Wenzhi Fang, Nicholas Tzou, Avinash Amballa, Yashas Malur Saidutta, Wenbo Li. On both 2026 cataloged papers Chappidi sits in the last author position without a corresponding-author flag, a pattern consistent with a senior or advisory role inside a small, stable team rather than a wide external collaboration web.
Activity67/100
Scores 67/100. OpenAlex counts 9 works in the last three years, with yearly totals of 2 (2020), 3 (2021), 1 (2022), 4 (2023), 4 (2024) and 5 (2026) - a dip in 2022 followed by a sustained rise, with 2026 the highest year on file, including both reviewed arXiv papers on LoRA fine-tuning and merging.
Reproducibility55/100
Scores 55/100, computed from Evolve reproducibility reviews of 2 papers (mean 55), both rated medium readiness. Strengths flagged by reviewers: principled constructions - freezing a null basis of reasoning activations into LoRA, and a data-free SVD-based Net Utility metric for merging. Limitations: small-scale, single-seed evidence and no variance reported on the +2.1/+2.2 average gains.
Research summary
Srinivas Chappidi is a machine learning researcher at Samsung (United States) whose Evolve-cataloged work centers on parameter-efficient fine-tuning (PEFT) of large language models. The profile is name-resolved from public bibliographic data (OpenAlex, identity field: openalex) and linked to ORCID 0009-0001-1313-4409. Two 2026 arXiv papers sit in the Evolve catalog, both with Chappidi in the last author position: a null-basis LoRA method that freezes a null basis of reasoning activations to preserve post-RL reasoning during supervised fine-tuning, and a budget-aware LoRA merging framework built on a data-free Net Utility metric that beats uniform rank allocation by +2.1/+2.2 average. Together they split his cataloged field distribution evenly (0.5 LLM fine-tuning, 0.5 model merging/PEFT). The broader OpenAlex record shows 9 works in the last three years, with yearly counts of 2 (2020), 3 (2021), 1 (2022), 4 (2023), 4 (2024) and 5 (2026), plus an h-index of 5 and i10-index of 3. Evolve's reproducibility reviews average 55 across the two papers, both rated medium readiness. One caveat: the OpenAlex topic profile also surfaces clusters in topic modeling, search behavior and geothermal energy systems that do not appear in the reviewed catalog, possibly reflecting name-disambiguation noise or earlier work in other domains.
Collaboration pattern
Chappidi publishes with a compact, stable team: 6 distinct coauthors, with the recurring names Lazar Valkov, Wenzhi Fang, Nicholas Tzou, Avinash Amballa, Yashas Malur Saidutta and Wenbo Li likely spanning both 2026 papers. He takes the last author position on each reviewed paper without a corresponding-author flag, consistent with a senior contributor or advisor role. The modest collaboration score (39) reflects this small network rather than inactivity.
Research keywords
Highlights
Null-basis LoRA preserves post-RL reasoning during fine-tuning
Data-free Net Utility ranks LoRA merges within a budget
Timeline
- 2020OpenAlex records the earliest works on file, with 2 publications in 2020.
- 2023Output recovers to 4 works in 2023, beginning a sustained climb after the 2022 dip (1 work).
- 2024A further 4 works in 2024, contributing to 9 works across the last three years.
- 2026Highest year on file: 5 works, including both reviewed arXiv papers - null-basis LoRA fine-tuning and budget-aware LoRA merging - each with Chappidi as last author.
Papers in the Evolve catalog
Frequent co-authors
About this profile
This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.
- Bibliometrics come from OpenAlex only; name disambiguation is uncertain - the topic profile includes geothermal-energy and search-behavior clusters absent from the reviewed catalog.
- Output and impact dimensions are unscored (insufficient data); only h-index 5 and i10-index 3 are on file, with no per-paper citation data.
- Reproducibility scores cover only the 2 reviewed papers (mean 55); the wider record is unreviewed, and both papers are arXiv preprints, not yet peer-reviewed venues.
- Affiliation is listed as Samsung (United States) with country code KR, taken as provided and not independently verified.