Researcher at Accenture (Switzerland) working on efficient LLM inference via layer skipping and conditional computation
2
papers here
1
reviewed by Evolve
2
first-author
2
corresponding
—
works · OpenAlex
0
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-17) 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—
Insufficient data: the snapshot shows an h-index and i10-index of 0 with only 4 recent works, so cumulative output volume, venue strength, and per-year production cannot be scored yet.
Impact—
Not yet measurable: the recorded h-index and i10-index are both 0 and no citation counts by year are provided, so this axis would measure citations and scholarly influence once they accrue.
Focus100/100
Focus scores a full 100: the 4 recorded works concentrate in tightly adjacent topics — Topic Modeling (4), Advanced Neural Network Applications (3), Natural Language Processing Techniques (3), and Domain Adaptation and Few-Shot Learning (2) — with the one classified field being efficient NLP/LLM inference via conditional computation.
Collaboration0/100
This axis scores 0: the snapshot records 0 distinct co-authors and no frequent co-author list, and both cataloged 2026 papers list this researcher as first and corresponding author, consistent with a predominantly solo working mode; partnership patterns cannot be characterized further from available data.
Activity47/100
Activity scores 47 from a thin but current record: 4 works in the most recent three-year window, all assigned to 2026 in the OpenAlex counts-by-year data, with h-index and i10-index both at 0. This is an early, front-loaded publication pattern rather than an established sustained cadence.
Reproducibility55/100
Reproducibility averages 55 across 1 reviewed paper. Evolve describes LayerRoute as transparent, with a verified but modest ~1.04x speedup on Qwen2.5-0.5B, though key controls are missing, and rates its repro-readiness medium. The second 2026 paper — a rigor-matched audit of periodic-step layer skipping comparing ConFlayers with SWIFT — is not yet reviewed, so this mean rests on a single data point and may shift as audits accumulate.
Research summary
Prateek Kumar Sikdar is an industry-affiliated researcher at Accenture (Switzerland) whose visible record centers on efficient large language model inference. The OpenAlex-resolved snapshot contains 4 works, all dated 2026, with h-index and i10-index both at 0 — an early, tightly concentrated profile. Topic tags cluster around Topic Modeling (4 works), Advanced Neural Network Applications (3), Natural Language Processing Techniques (3), and Domain Adaptation and Few-Shot Learning (2), and the single field classification maps to efficient NLP/LLM inference via conditional computation. Two 2026 arXiv preprints appear in the Evolve catalog, each with this researcher as first and corresponding author. LayerRoute introduces per-layer hard gates combined with LoRA on Qwen2.5-0.5B; Evolve's review verified a modest ~1.04x speedup and called the method transparent while noting missing key controls, rating repro-readiness medium. A companion audit rigor-matches periodic-step layer skipping, comparing ConFlayers with SWIFT and adding a supplemental analysis of trained routing alternatives; it has not yet been reviewed. The snapshot lists 0 distinct co-authors, pointing to solo-led work, though the absence of an ORCID link means co-authorship could be under-counted. Overall this is a nascent but sharply focused line of research in conditional computation, best read as an early-stage profile that will firm up as citations and reviews accumulate.
Collaboration pattern
The snapshot records 0 distinct co-authors and no frequent co-authors; both cataloged 2026 papers list this researcher as first and corresponding author, indicating solo-led work. Because identity is resolved from OpenAlex without an ORCID link, co-authorship and affiliation matching may be incomplete, so this pattern should be read as preliminary.
Research keywords
Highlights
LayerRoute: adaptive layer skipping with LoRA-preserved quality
Rigor-matched audit of periodic-step layer skipping
A second 2026 first-authored preprint rigor-matches periodic-step layer skipping for efficient LLM inference, comparing ConFlayers with SWIFT and adding a supplemental analysis of trained routing alternatives; Evolve has not yet reviewed it, so no verdict or repro-readiness rating exists.
Timeline
- 2026Published LayerRoute on arXiv — adaptive layer-skipping with LoRA-preserved quality for efficient LLM inference — as first and corresponding author.
- 2026Published a rigor-matched audit of periodic-step layer skipping (ConFlayers versus SWIFT), with a supplemental analysis of trained routing alternatives.
- 2026OpenAlex snapshot recorded 4 works in the most recent three-year window, all dated 2026, with h-index and i10-index at 0.
Papers in the Evolve catalog
- first authorcorresponding
LayerRoute: Adaptive Layer-Skipping with LoRA-Preserved Quality for Efficient LLM Inference
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.
- Identity is name-resolved from public OpenAlex bibliographic data; no ORCID is linked, so the record may merge or omit works by namesakes.
- All bibliometrics (h-index 0, i10-index 0, 4 recent works, topic and field distributions) come from a single OpenAlex snapshot.
- The Evolve catalog holds only two 2026 arXiv preprints; the second lacks a review verdict, keywords, and a repro-readiness rating.
- countsByYear contains only 2026 (4 works), so no earlier publication history is visible in this snapshot.