Early-career ML researcher at North Carolina State working on LLM token efficiency and adversarial robustness
3
papers here
2
reviewed by Evolve
3
first-author
0
corresponding
—
works · OpenAlex
1
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—
Not yet scored: indexing coverage is too thin for a reliable output measure at this career stage. The axis would track total works and yearly throughput once coverage matures; h-index 1 and i10-index 0 alone are not a fair basis.
Impact—
Not yet scored: with an i10-index of 0, citation-based impact cannot be meaningfully measured yet. This axis would aggregate citations, citing works and influence beyond the author's field; arXiv-only preprints typically accrue citations later, so revisit.
Focus0/100
Focus scores 0 because the profile is unusually dispersed for its stage: the field distribution splits evenly at 0.5 between LLM efficiency and interpretability and adversarial robustness, and topic tagging spreads across five clusters (adversarial robustness 5, topic modeling 4, neural-network applications 3, privacy-preserving 2, graph neural networks 2). Broad exploration, not yet one centered line.
Collaboration46/100
Collaboration scores 46 on a modest base: 8 works in the recent three-year window with 9 distinct co-authors, roughly one new collaborator per paper. All three catalog papers are first-authored and none is corresponding. Recurring names include Ferhat Erata, Joe Lilien, MohamadAli Torkamani, Lingdao Sha, Xueyu Mao, Yang Liu, Peijie Qiu, Rui Song and Xiaorui Liu.
Activity64/100
Activity scores 64, driven by a clear acceleration: yearly output went 1 (2023), 2 (2024), 1 (2025), then 5 in 2026, with 8 works in the recent three-year window. The 2026 spike includes three first-author arXiv preprints in the Evolve catalog, suggesting the current rate is this researcher's most productive period so far.
Reproducibility38/100
Reproducibility scores 38 as the mean across the 2 scored papers (reproMean 38). TEFM is rated low readiness — results rest on a single run and no code was released — while the budget-conditioned robustness paper (TCN) is rated medium. Neither paper with an Evolve verdict reports released code, so independent verification currently depends on author-shared artifacts.
Research summary
Zhichao Hou is a machine-learning researcher at North Carolina State University whose early record already spans two distinct fronts. The first is LLM efficiency and interpretability, where the first-author 2026 preprint TEFM compresses structured records into three RQ-VAE codes and reportedly cuts LLM input by about 99% with under 2% accuracy loss. The second is adversarial robustness, where the TCN preprint conditions one network on a continuum of perturbation budgets through Fourier-FiLM, delivering efficient, strong PGD gains while, per Evolve's review, somewhat overclaiming the all-budgets claim on AA and Tiny-ImageNet. A third first-author preprint applies stratified consistency distillation to natural language formalization and awaits Evolve review. Bibliometrically the profile is early-stage: OpenAlex attributes an h-index of 1 and an i10-index of 0, with 8 works in the recent three-year window and a clear 2026 acceleration (1, 2, 1, then 5 works by year). Topic tagging shows breadth — adversarial robustness (5), topic modeling (4), neural-network applications (3), privacy-preserving technologies (2), graph neural networks (2) — and the field distribution splits evenly between LLM efficiency/interpretability and adversarial robustness. Collaboration is modest but broadening, with 9 distinct co-authors, and reproducibility scoring across 2 papers averages 38, held back by single-run results and unreleased code.
Collaboration pattern
All three catalog papers are first-authored and none lists corresponding authorship. OpenAlex counts 9 distinct co-authors across 8 recent works — roughly one new collaborator per paper. Recurring names include Ferhat Erata, Joe Lilien, MohamadAli Torkamani, Lingdao Sha, Xueyu Mao, Yang Liu, Peijie Qiu, Rui Song and Xiaorui Liu, indicating a steadily widening but still small network.
Research keywords
Highlights
TEFM: ~99% LLM input reduction on structured data via 3 RQ-VAE codes
One model, many threat levels: budget-conditioned adversarial robustness
Stratified consistency distillation for natural language formalization
A third first-author 2026 arXiv preprint applying stratified consistency distillation to natural language formalization. Evolve has not yet published a review verdict, keywords or reproducibility rating for this paper.
Timeline
- 20231 indexed work recorded in the OpenAlex yearly counts, the earliest point of this profile.
- 20242 indexed works recorded in the OpenAlex yearly counts.
- 20251 indexed work recorded, a dip before the 2026 acceleration.
- 20265 works recorded; three first-author arXiv preprints enter the Evolve catalog — TEFM, one-for-all robustness, and stratified consistency distillation.
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.
- All bibliometrics come from OpenAlex only; other indexes such as Scopus or Google Scholar were not consulted.
- Output and impact dimensions are missing: h-index is 1 and i10-index is 0, typical for an arXiv-heavy early-career profile where citations lag.
- The stratified consistency distillation paper has no Evolve verdict, keywords, slug or reproducibility rating yet.
- The profile is name-resolved from public bibliographic data (OpenAlex record plus ORCID 0009-0003-9894-1115); identity details beyond this were not verified.