Our Work

What's Changed Dev by @jcvdav in https://github.com/human-ocean/PNA_HS_closure/pull/3 Full Changelog: https://github.com/human-ocean/PNA_HS_closure/compare/v0.0.1...v0.0.2

Este trabajo presenta una investigación sobre el efecto del glifosato en Drosophila melanogaster: En la mayoría de los organismos con reproducción sexual, los mecanismos involucra…

Este trabajo presenta una investigación sobre el efecto del glifosato en Drosophila melanogaster: En la mayoría de los organismos con reproducción sexual, los mecanismos involucrados en la determinación del sexo conducen a la obtención de un 50% de hembras y de machos, aproximadamente. Sin embargo, esta respuesta puede modificarse por la exposición a sustancias conocidas como disruptores endócrinos. Un compuesto de interés es el herbicida glifosato, su persistencia en suelos oscila de 2 a 174 días posteriores a su aplicación, por lo que se requiere evaluar sus efectos en exposiciones prolongadas y continuas. El objetivo de este trabajo fue evaluar en Drosophila melanogaster el efecto de la exposición a glifosato, a lo largo de catorce generaciones, utilizando como biomarcador, la proporción sexual. Larvas de tercer estadio de la cepa Canton-s se expusieron mediante alimentación a ocho concentraciones diferentes de glifosato (Faena©) y a agua destilada como testigo negativo. Una vez que emergieron los adultos se tomaron al azar diez parejas provenientes del testigo y de cada concentración. Cada pareja se sembró en un tubo con medio de cultivo para obtener su progenie y registrar el número de hembras y machos con el fin de calcular la proporción sexual. El resto de las moscas tratadas se utilizaron para obtener larvas que fueron expuestas a las mismas concentraciones de glifosato que sus progenitores. Este proceso se repitió a lo largo de catorce generaciones. Los resultados muestran que en Drosophila el glifosato modificó la proporción sexual de las moscas expuestas durante catorce generaciones, por lo que se concluye que es un posible disruptor endócrino.  

The file contains the data which was used to generate the figures in the manuscript.

The transparent open box (TOB) learning network algorithm offers an alternative approach to the lack of transparency provided by most machine-learning algorithms. It provides the exact calculations an…

The transparent open box (TOB) learning network algorithm offers an alternative approach to the lack of transparency provided by most machine-learning algorithms. It provides the exact calculations and relationships among the underlying input variables of the datasets to which it is applied. It also has the capability to achieve credible and auditable levels of prediction accuracy to complex, non-linear datasets, typical of those encountered in the oil and gas sector, highlighting the potential for underfitting and overfitting. The algorithm is applied here to predict bubble-point pressure from a published PVT dataset of 166 data records involving four easy-to-measure variables (reservoir temperature, gas-oil ratio, oil gravity, gas density relative to air) with uneven, and in parts, sparse data coverage. The TOB network demonstrates high-prediction accuracy for this complex system, although it predictions applied to the full dataset are outperformed by an artificial neural network (ANN). However, the performance of the TOB algorithm reveals the risk of overfitting in the sparse areas of the dataset and achieves a prediction performance that matches the ANN algorithm where the underlying data population is adequate. The high levels of transparency and its inhibitions to overfitting enable the TOB learning network to provide complementary information about the underlying dataset to that provided by traditional machine learning algorithms. This makes them suitable for application in parallel with neural-network algorithms, to overcome their black-box tendencies, and for benchmarking the prediction performance of other machine learning algorithms.

Purpose: This study explores how Artificial Intelligence (AI) has emerged as a tool for measuring human intelligence and ascertains whether the AI-based methods in measurement can supplement, enhance …

Purpose: This study explores how Artificial Intelligence (AI) has emerged as a tool for measuring human intelligence and ascertains whether the AI-based methods in measurement can supplement, enhance or substitute psychometric evaluation in any way. The study will also explore how the fields of psychometrics and AI are converging, highlight the essential AI-based methods in intelligence assessment, point out their strengths and weaknesses, and discuss the issues of fairness, impartiality, validation, and responsibility in relation to them. Method/Approach: The research employs a qualitative, theoretical, and interpretative methodology that draws on the narrative-thematic assessment of secondary literature sources. The pertinent articles from peer-reviewed articles and preprints along with institutional reports were subjected to thematic content analysis. The literature was categorised into theme groups as follows: classical psychometric principles, AI-based computerised adaptive testing, brain imaging-based prediction, behaviour and multi-modal signal analysis, AI processes used in the human intelligence tests, and issues of fairness and ethics. Results/Findings: The results indicate that AI has been shown to be the most effective tool for enhancing the efficiency and accuracy of standard intelligence testing methods in terms of computerized adaptive testing and automated scoring. Methods based on neuroimaging provide predictions about intelligence that are statistically weak but scientifically valid; behavioral methods such as eye-tracking have proven to be promising but may require substantial resources to be applied effectively. Yet, there is no single AI method that can provide full construct validity, standardization and reliability, and fairness as compared to conventional intelligence tests administered by humans. Moreover, these results once again bring to light issues of algorithmic bias, consent, transparency, and instability of responses in generative AI methods. Originality/Value: This research offers a coherent framework for comparing research studies on different aspects of AI in psychometrics, adaptive testing, neuroimaging, behavioral inference, and evaluation of AI technology. It presents a hybrid assessment model involving AI in adaptive delivery, scoring, and signal extraction while allowing qualified professionals to stay in charge of interpreting results and ensuring ethical integrity. Type of Paper: Exploratory Research.

Conditional mathematical arguments need careful bookkeeping: which obligations were incurred, which have been addressed, and which were addressed in a way that actually proves the target. Audited Oper…

Conditional mathematical arguments need careful bookkeeping: which obligations were incurred, which have been addressed, and which were addressed in a way that actually proves the target. Audited Operational Realisability (AOR) is a mathematical theory of this bookkeeping. An audit state consists of demanded records and stored certificates; each certificate carries a status, such as budgeted or blocked, and a payload whose meaning is fixed in advance. We separate three conclusions that are easily conflated. Inference closure registers every dependency of a demanded record, accounting requires a stored certificate for each demanded or dependency-reachable record, and delivery requires a certificate whose status an explicit consumer policy accepts as proving its target. A certificate recording that a budget is exceeded accounts for the obligation but does not deliver feasibility to a consumer who requires it. We prove that inference closure is a reflection that creates no evidence, in a category whose arrows preserve certificates exactly, and that this category has colimits in which every accepted certificate is already stored at some object of the diagram. For operations that add evidence, we prove finite completion when every unaccounted state satisfying an invariant admits a legal step that preserves the invariant and lowers a natural-number rank, and confluence under replayability, and we strictly separate accounting, recoverability, and uniform quantitative control. Retained evidence passes to increasing unions, and witnesses pass to limits under compactness and closed validity, provided an accepted payload is returned at the limit; without compactness they may escape. Two examples mark the boundary of accounting: a query system that pays its declared costs, is accounted, and identifies each hidden world along its exhaustive run, yet admits no finite-stage guarantee uniform over all worlds; and a target on binary streams, Lipschitz for the first-difference metric, whose shrinking finite-prefix certificates, priced within a summable credit, are all accepted, with centers converging to the target. The numbered theorems and the lemma are formalized in Lean 4 with Mathlib; the analytic estimates of applications remain explicit hypotheses.

Suppose a vector is observed through two families of linear measurements: a native family that we already control and a target family that we would like to control. How much of the target response is …

Suppose a vector is observed through two families of linear measurements: a native family that we already control and a target family that we would like to control. How much of the target response is left unexplained by the native measurements? We answer this question on finite-dimensional real or complex Hilbert spaces carrying a positive semidefinite energy. The answer is the adequacy residual, a generalized Schur complement whose quadratic form gives, for each combination of target measurements, the largest squared response at energy at most one that is invisible to the native family. We show that the residual is the smallest error left by any linear prediction of the target from the native measurements, that it vanishes exactly when the target factors through the native family, and that it obeys a Schur chain rule when measurements are added; these identities are classical facts about generalized Schur complements in the present notation. We then give explicit, checkable conditions for three practical tasks: shrinking the residual by a fixed fraction with added measurements, deciding exactly which directions an isometric restriction of the space must keep, and recovering an estimate after it has been transported to another space. An appendix treats sequences of such estimates. For a bounded native map L on a real Hilbert space and the identity as target, we take the residual to be projection onto the kernel of L; in finite dimensions this agrees with the Schur residual, and in any dimension a state v whose native part has squared norm at most K and whose residual part has squared norm at most q‖v‖2 with q<1 satisfies ‖v‖2 ≤ K/(1 − q). Energies may be singular, provided the measurements vanish on zero-energy vectors. Every numbered result is formalized in Lean 4 with Mathlib.