Overfitting remains a critical challenge in data-driven financial modeling, where machine learning (ML) systems learn spurious patterns in historical prices and fail out of sample and in deployment. This paper introduces the GT-Score, a composite objective function that integrates performance, statistical significance, consistency, and downside risk to guide optimization toward more robust trading strategies. This approach directly addresses critical pitfalls in quantitative strategy development, specifically data snooping during optimization and the unreliability of statistical inference under non-normal return distributions. Using historical stock data for 50 S P 500 companies spanning 2010-2024, we conduct an empirical evaluation that includes walk-forward validation with nine sequential time splits and a Monte Carlo study with 15 random seeds across three trading strategies. In walk-forward validation, GT-Score improves the generalization ratio (validation return divided by training return) by 98
IntroductionOverfitting remains a persistent barrier to reliable machine learning, especially in modern overparameterized deep models.MethodsWe conducted a narrative review synthesizing approximately 95 core studies (1943–2026) identified through structured searches of IEEE Xplore, ACM Digital Library, arXiv, Google Scholar, and Semantic Scholar, extracting mechanisms, assumptions, and empirical evidence for prominent methods for mitigating overfitting.ResultsWe organize techniques into a unified five-family taxonomy (parameter-, training-, data-, ensemble-, and objective-based) and provide a practical decision framework that maps data regimes, model families, and real-world scenarios to actionable regularization strategies.ConclusionOverfitting mitigation benefits from coordinated choices in data, model capacity, optimization, and evaluation. Our taxonomy and decision framework help practitioners select complementary interventions and avoid common pitfalls such as leakage and over-regularization.
Large Language Models appear competent when answering general questions but often fail when pushed into domain-specific details. No existing methodology provides an out-of-the-box solution for measuring how deeply LLMs can sustain accurate responses under adaptive follow-up questioning across arbitrary domains. We present DepthCharge, a domain-agnostic framework that measures knowledge depth through three innovations: adaptive probing that generates follow-up questions based on concepts the model actually mentions, on-demand fact verification from authoritative sources, and survival statistics with constant sample sizes at every depth level. The framework can be deployed on any knowledge domain with publicly verifiable facts, without requiring pre-constructed test sets or domain-specific expertise. DepthCharge results are relative to the evaluator model used for answer checking, making the framework a tool for comparative evaluation rather than absolute accuracy certification. Empirical validation across four diverse domains (Medicine, Constitutional Law, Ancient Rome, and Quantum Computing) with five frontier models demonstrates that DepthCharge reveals depth-dependent performance variation hidden by standard benchmarks. Expected Valid Depth (EVD) ranges from 3.45 to 7.55 across model-domain combinations, and model rankings vary substantially by domain, with no single model dominating all areas. Cost-performance analysis further reveals that expensive models do not always achieve deeper knowledge, suggesting that domain-specific evaluation is more informative than aggregate benchmarks for model selection in professional applications.
The lived experiences of children with neurodevelopmental and learning disabilities are documented through diverse forms of personal data, including identity, medical, mental health, accomplishments, social relationships, financial resources, and personal ecology. However, these records rarely integrate quality of life (QoL) information represented through validated multidimensional frameworks. Building on the gap identified in a companion narrative review, this paper presents a reproducible ontology engineering methodology for representing QoL in children with neurodevelopmental and learning disabilities. The paper presents a design-science methodology for constructing an ontology from the Self Research Institute's Self Data Atlas (SDA), a seven-category taxonomy for organizing longitudinal personal data. Derived from Wieringa's design-science framework, the methodology follows a seven-step protocol comprising: (1) competency-question-based requirements specification; (2) ontology reuse and re-engineering; (3) conceptualization; (4) crisp formalization in OWL 2 DL; (5) fuzzy formalization in Fuzzy OWL 2; (6) alignment with the Common Core Ontology (CCO) as a Basic Formal Ontology (BFO)-conformant mid-level scaffold; and (7) iterative evaluation and refinement. The methodology incorporates two complementary design features. First, it separates factual and classificatory information, including demographics, SNOMED CT diagnostic codes, and clinical observations, represented in classical description logic, from graded and context-dependent QoL concepts, including social inclusion, severity, intensity, and wellbeing, represented using fuzzy datatypes, fuzzy linguistic variables, and graded class membership. Second, it embeds tritemporal reasoning by assigning three temporal qualifiers to each SDA event, state, or information artifact: (1) absolute time, representing when the fact existed in reality; (2) ingestion time, representing when the record entered the system; and (3) assignment time, representing when an interpretative classification was applied. These temporal qualifiers are represented as ontology entities rather than opaque metadata. This tritemporal structure supports longitudinal queries, retrospective recording, and event-history replay. Each stage is illustrated using examples from the seven SDA categories. The paper illustrates how CCO classes provide stable upper-level abstractions for SDA components and interoperability with ICF, SNOMED CT, HL7 FHIR, and the Schalock–Verdugo QoL model. This paper makes a methodological rather than an empirical contribution. It provides a reproducible ontology engineering methodology for future SDA ontology development and related personal-data ontologies.
This study examines the pivotal, yet frequently overlooked, concept of risk velocity, the speed at which risk repercussions unfold after an initial trigger. It highlights how existing risk management, focused mainly on probability and impact, is inadequate for handling rapidly escalating events, as seen in crises like the COVID-19 pandemic, the CrowdStrike–Microsoft outage, the Change Healthcare ransomware attack, and the Aflac data breach. These incidents reveal how swift-moving threats can outpace traditional response mechanisms, causing severe operational and reputational harm. In today's fast-evolving threat landscape, technologies such as AI, AGI, ML, deep learning, and autonomous response act as risk multipliers and critical mitigation tools. AI-driven threats accelerate incident velocity but, conversely, enable faster detection and automated response. Building on models like Time-to-Impact Analysis and Agility and Resilience Theory, this chapter advocates for incorporating risk velocity alongside probability and impact, with velocity-aware matrices and AI-enhanced simulations.
