Johnson Ang

Mr. Johnson Ang

orcid

Research Focus & Metrics

Academic Career Length

1Year (2026-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
28 September 2026 in SSRN Electronic Journal
Verified
Preprint
Open access
JAJohnson Ang
Johnson Ang
Latent interpolation is often judged by whether generated images appear smooth, but visual smoothness and latent-space geometry are not the same object. This paper critically evaluates four traversal strategies in a pretrained generative adversarial network (GAN): linear interpolation (LERP), spherical linear interpolation (SLERP), a stochastic random walk, and a deliberately perturbed quadratic path. The pilot implementation measures latent-vector norms, cross-method structural similarity (SSIM), Laplacian-variance sharpness, principal-component projections, and qualitative image continuity. For the sampled LERP/SLERP trajectory, latent-norm variance is approximately 1.3563 for LERP and 0.3111 for SLERP, while matched-step LERP/SLERP images remain structurally similar (mean SSIM approximately 0.9344; minimum approximately 0.8248). These observations demonstrate a useful separation: a substantial radial difference in latent space need not induce a comparably large structural difference in generated images. The paper also identifies two common interpretive errors. First, SLERP is a geodesic on a sphere under a spherical metric, not automatically a geodesic of the generator-induced image manifold. Second, cross-method SSIM and Laplacian variance do not by themselves measure perceptual trajectory smoothness or realism. A repeatedpairs protocol is therefore specified for confirmatory follow-up, including consecutive-frame perceptual distance, path-length controls, multiple seeds, and paired statistical analysis. The contribution is a reproducible framework for distinguishing geometric regularity from perceptual continuity without overstating what a single latent trajectory can establish.

17 September 2026 in Preprints.org
Verified
Preprint
Open access
PDF file
JAJohnson Ang
Johnson Ang
Production AI systems are socio-technical systems whose behavior emerges from interacting states across data, code, execution environments, model or provider state, prompts, retrieval context, tools, policies, evaluators, infrastructure, and human intervention. This paper develops an evidence-centric framework that treats Machine Learning Operations (MLOps) and Large Language Model Operations (LLMOps) as disciplines for controlled change in such systems. The framework integrates evidence from recent MLOps architecture studies, LLMOps research, generative-AI evaluation literature, current platform documentation, security guidance, and AI governance frameworks. Five contributions are developed: (1) a multidimensional architecture lens that extends infrastructure-friction versus operational-yield analysis into an explicit trade-off space; (2) executable provenance, which defines reproducibility as reconstruction of materially causal execution state; (3) a composable MLOps/LLMOps architecture separating evidence, evaluation, orchestration, compute, knowledge, serving, telemetry, policy, and human control; (4) evaluation as a release-control mechanism in which the evaluator is itself versioned; and (5) a capability-coverage model for organizational readiness. The paper derives eight falsifiable hypotheses and specifies a controlled benchmark protocol spanning supervised ML, parameter-efficient fine-tuning, retrieval-augmented generation (RAG), and tool-using agents. An author-supplied Kubeflow-oriented repository is examined as an implementation artifact and is explicitly distinguished from evidence about current Kubeflow Pipelines v2 behavior. The resulting framework provides measurable constructs for studying architecture, provenance, evaluation, security, governance, organizational capability, and production economics without assuming a universal platform configuration.

5 September 2026 in SocArXiv
Verified
Preprint
Open access
JAJohnson Ang
Johnson Ang
This study presents a hybrid demographic-housing forecasting pipeline for preschool infrastructure demand in Singapore, designed for the Early Childhood Development Agency (ECDA). The model integrates cohort-component population aging with Build-To-Order (BTO) housing delivery signals and Census 2020 subzone-specific child-per-unit ratios, replacing fixed national coefficients that introduce up to 10% estimation error (Wilson et al., 2021). OLS regression models the population-to-child-count relationship (R^2=0.87), while Monte Carlo simulation (n=1,000) transforms deterministic point-estimates into probabilistic planning ranges, establishing a 95th percentile confidence bound recommending a three-center national safety buffer. The pipeline embeds algorithmic governance tripwires — automated buffer-consumption thresholds at 25%, 50%, and 100% — that mandate escalating executive intervention before fiscal floors are breached (NIST, 2023). A Social Return on Investment (SROI) reconciliation maps pedagogical outcomes across five weighted domains to the Heckman Equation, adopting a conservative 1.75x social return estimate (Heckman et al., 2010). Break-even analysis identifies a 3.5-month fiscal inflection point with a 4.62-month margin of safety for construction acceleration investments.Keywords: Infrastructure Planning, Early Childhood Education, Demographic Forecasting, Algorithmic Governance, Monte Carlo Simulation, Urban Analytics, Singapore Housing