Cheng-En Ou

—

Unclaimed · AI-structured profileOpenAlex

Early-career researcher working on RL-based battery storage dispatch and 4-bit quantization for LLM edge deployment

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-09) 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 scored yet: OpenAlex indexes only two works, both dated 2026, with h-index and i10-index recorded as 0, so productivity cannot be characterized. This axis would measure publication volume, venue mix, and authorship-role consistency over time.

Impact—

Not scored: no citation data is available and the only two indexed works both appeared in 2026, far too recent for citations to accumulate. This axis would measure citations, h-index trajectory, and downstream uptake of the two preprints.

Focus0/100

The focus score is 0, reflecting an evenly split profile: the field distribution allocates 0.5 to reinforcement learning for power systems/energy storage and 0.5 to LLM efficiency and quantization, while OpenAlex assigns five topic tags — topic modeling, advanced neural network applications, microgrid control and optimization, electric power system optimization, and smart grid energy management — each with a count of 1. No single theme yet dominates the two indexed works.

Collaboration22/100

Collaboration scores 22, low: only 2 distinct coauthors (Kuanlin Chen and Chen-Wei Kuo) appear across the two indexed works, and Cheng-En Ou sits in the last author position on both catalog papers without corresponding-author designation. This points to early-stage work inside one or two small teams rather than an extended co-authorship network.

Activity32/100

Activity scores 32, modest: OpenAlex records 2 works in the recent three-year window, both concentrated in a single year (countsByYear lists one entry of 2 for 2026) with no earlier indexed output. Both are arXiv preprints released in the same year, indicating a recent start rather than sustained multi-year production.

Reproducibility55/100

Reproducibility averages 55 across the 2 reviewed papers, both rated medium readiness. The PPO battery-dispatch paper is judged honest in its claims but thin on statistics, whereas the NF4 quantization study reports concrete correlation figures (ρs ≥ 0.980 for global ranks; 0.619–0.793 for top-k sets) across three small LLMs, supporting partial replication.

Research summary

Cheng-En Ou is an early-stage researcher whose indexed work to date spans two applied machine-learning directions: reinforcement learning (RL) for power systems and energy storage, and LLM efficiency and quantization. OpenAlex currently indexes two works, both dated 2026, and the Evolve catalog reviews both as arXiv preprints. The first proposes cost-aware RL with layered action masking and projection, combined with an advisory PPO setup, for battery energy storage dispatch under suppressed-spread market shifts; Evolve's review calls the feasibility claims honest but notes PPO loses to an MPC baseline, profits are trivially small, and statistics are thin. The second is an empirical quantization study showing that 4-bit NF4 weight-only quantization preserves global BOS-sink attention-head ranks (ρs ≥ 0.980) while top-k head sets shift substantially (0.619–0.793) across three small LLMs, motivated by edge deployment on Jetson Orin NX hardware. Bibliometrically the profile is thin: h-index and i10-index are recorded as 0, recent three-year works number 2, and only 2 distinct coauthors appear (Kuanlin Chen and Chen-Wei Kuo). The OpenAlex field distribution splits evenly, 0.5 to each direction, and five topic tags each carry a count of 1. Both cataloged papers receive a medium reproducibility-readiness rating, averaging 55.

Collaboration pattern

Collaboration is narrow: 2 distinct coauthors — Kuanlin Chen and Chen-Wei Kuo — appear across the two 2026 preprints, and Cheng-En Ou sits in the last author position on both, without corresponding-author designation. This is consistent with small-team, early-stage work within one or two groups rather than a wide co-authorship network; expanding beyond this pair and taking corresponding roles would strengthen the collaboration profile.

Research keywords

battery energy storage dispatchelectricity market arbitragesafe reinforcement learningPPOaction maskingattention sinkpost-training quantizationNF4 weight-only quantizationcalibration transferedge deploymentJetson Orin NXLLM efficiency

Highlights

Safe RL for battery storage dispatch under collapsing spreads
Proposes layered action masking and projection with an advisory PPO agent for cost-aware battery energy storage dispatch; Evolve's review finds the feasibility claims honest under suppressed-spread market shifts, but notes PPO loses to an MPC baseline, profits are trivially small, and statistics are thin.
4-bit NF4 quantization preserves global BOS-sink head ranks
An empirical study across three small LLMs showing that 4-bit NF4 weight-only quantization keeps global BOS-sink attention-head ranks stable (ρs ≥ 0.980) even as top-k head sets shift (0.619–0.793), with implications for edge deployment on Jetson Orin NX hardware.
A dual-track applied machine-learning portfolio
The two 2026 preprints split evenly between safe RL for energy storage and LLM quantization for edge hardware, matching the 50/50 OpenAlex field distribution and sharing a medium reproducibility-readiness rating with an average of 55 across both papers.

Timeline

2026Releases an arXiv preprint on cost-aware reinforcement learning combining action masking, projection, and advisory PPO for battery energy storage dispatch under suppressed-spread market shifts.
2026Releases an arXiv preprint showing that 4-bit NF4 weight-only quantization preserves global BOS-sink head ranks (ρs ≥ 0.980) across three small LLMs, while top-k sets shift.
2026Current snapshot: OpenAlex indexes 2 works, both dated 2026, with h-index and i10-index recorded as 0 and 2 distinct coauthors in the network.

Papers in the Evolve catalog

Frequent co-authors

Kuanlin Chen×2 Chen-Wei Kuo×2

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.

Bibliometric signals are computed from OpenAlex records only; other databases (Scopus, Web of Science, Semantic Scholar) were not consulted.
No verified institution, country, or ORCID is attached; the profile is name-resolved from public bibliographic data and may merge or split identities for common names.
Output and impact dimensions cannot be scored: only 2 works are indexed (both 2026), with h-index and i10-index recorded as 0.
Both reviewed papers are arXiv preprints; peer-review status and venue acceptance are not captured in the available data.