DIGITUM Repositorio Institucional de la Universidad de Murcia
Not a member yet
77160 research outputs found
Sort by
Predicción sensible al contexto precisa para técnicas especulativas en procesadores
No full textCon cada nueva generación, los procesadores de alto rendimiento adoptan técnicas cada vez más agresivas para extraer más rendimiento. La ejecución fuera de orden, los mecanismos avanzados de comunicación de memoria y la fusión de instrucciones aportan mejoras significativas, pero dependen en gran medida de la especulación. A medida que los diseños se amplían, estos mecanismos especulativos se vuelven más vulnerables a los fallos de predicción, lo que puede costar hasta un 15% del rendimiento potencial.
En este contexto, la predicción de dependencias de memoria es esencial para garantizar que los loads no pasen por alto los stores más antiguos a la misma dirección. Asimismo, la fusión de instrucciones se ha utilizado para reducir la presión sobre el pipeline, aunque las implementaciones industriales se centran casi exclusivamente en patrones consecutivos simples. Solo una propuesta previa exploró la fusión especulativa de instrucciones de memoria no consecutivas basada en predicción.
Sin embargo, los enfoques actuales para la predicción de dependencias de memoria y la fusión especulativa siguen siendo limitados. La mayoría adoptan ideas procedentes de predictores de saltos, pese a que, como defendemos en esta tesis, no son adecuadas para estos dominios y suelen producir ineficiencias. Sostenemos que la información contextual utilizada para las predicciones debe seleccionarse con mayor cuidado. El uso de historias de longitud fija, o la búsqueda por fuerza bruta de la mejor longitud, puede generar rutas demasiado largas que introducen ruido o demasiado cortas que producen aliasing.
Defendemos que cada predicción debe entrenarse únicamente con la información situada entre las instrucciones implicadas. Si el camino entre dos instrucciones se repite, también es probable que se repita su interacción.
Bajo esta premisa, el objetivo de la tesis es revisar mecanismos especulativos desde la perspectiva de la sensibilidad al contexto. La hipótesis es que la precisión y la eficiencia pueden mejorar significativamente si los mecanismos se adaptan al contexto dinámico específico de cada interacción, en lugar de basarse en historias estáticas o genéricas. Para validar esta hipótesis, diseñamos y evaluamos nuevos predictores capaces de aprovechar el contexto relevante para ampliar la especulación y reducir sus riesgos.
La primera parte de la tesis estudia al Store Sets, predictor que continúa utilizándose como referencia en la predicción de dependencias de memoria. Nuestro análisis identifica cinco limitaciones fundamentales, en su mayoría relacionadas con la serialización innecesaria de instrucciones de memoria.
Basándonos en estas observaciones, presentamos PHAST, un nuevo predictor de dependencias que abandona los enfoques de distancia fija y de conjuntos. PHAST aprende de la ruta real de ejecución entre un store conflictivo y el load dependiente, permitiéndole identificar qué store debe proporcionar el valor al load con gran precisión. Este entrenamiento elimina numerosas falsas dependencias y reduce los squashes del pipeline. Con solo 14,5KB de presupuesto hardware, PHAST reduce el MPKI un 62% de media y obtiene un speedup del 1,29% respecto al estado del arte, con mejoras de hasta el 22%, acortando la distancia hasta un predictor ideal en apenas un 1,5%.
La última parte aborda la predicción de fusión de instrucciones. La única propuesta previa de fusión no consecutiva se fundamenta en un predictor Tournament, diseño que consideramos inadecuado para esta tarea. Para resolver sus limitaciones, presentamos FLIP, un predictor de fusión basado en el mismo principio de sensibilidad al contexto que PHAST. FLIP identifica oportunidades de fusión no consecutiva entre loads y stores, e introduce un conjunto de optimizaciones destinadas a relajar restricciones innecesarias y reducir los casos perjudiciales. Nuestros experimentos muestran que FLIP mejora el rendimiento en un 2,44% en SPEC CPU 2017 y en un 2,94% en MiBench, mientras reduce el MPKI en un 83%.With each new generation, high-performance processors adopt increasingly aggressive techniques to extract more performance. Out-of-order execution, advanced memory communication mechanisms, and instruction fusion all contribute to higher performance, but they rely heavily on speculation. As designs grow wider and deeper, these speculative mechanisms become more vulnerable to misprediction, which can cost up to 15% of the processor's performance potential.
In this context, memory dependence prediction is essential to ensure that loads do not bypass older stores to the same address. Similarly, instruction fusion has been used to reduce pressure on the pipeline, though current industrial implementations focus almost exclusively on simple, consecutive patterns. Only a single prior proposal has explored speculative fusion of non-consecutive memory instructions using a predictor.
However, state-of-the-art approaches in both memory dependence prediction and instruction fusion prediction remain limited. Most are directly inspired by branch predictor designs, which, as we claim in this thesis, are not well suited for these domains and often result in inefficiencies.
The central claim of this thesis is that context information must be carefully selected for accurate prediction. The use of using fixed lengths or finding the best suited among a set through brute-force, like how it is done in branch prediction, can lead to an explosion of paths when the history is larger than necessary, or to aliasing when shorter.
Instead, we state that each prediction must be trained only with the information between the instructions involved in it. If the same path between those two instructions recur, then the outcome of the prediction will likely recur too.
Following this premise, the objective of this thesis is to revisit speculative hardware mechanisms from the perspective of context sensitivity. Our working hypothesis is that prediction accuracy and efficiency can be substantially improved if mechanisms adapt to the specific dynamic context of each case, rather than relying on static or generic lengths. To test this hypothesis, we design and evaluate new predictors that exploit the relevant execution context to expand the reach of speculation while minimizing its risks.
The first part of this thesis revisits the Store Sets predictor, still widely used as a reference mechanism for memory dependence prediction. Our analysis identifies five key limitations, most rooted in the unnecessary serialization of memory instructions.
Building on these insights, we introduce PHAST, a new memory dependence predictor that departs from fixed-distance and set-based approaches. PHAST learns from the actual execution path between a conflicting store and its dependent load, enabling it to predict precisely the store from which the load should receive its value. This path-sensitive training eliminates many false dependences and significantly reduces pipeline squashes. With only 14.5 KB of hardware budget, PHAST reduces MPKI by 62% on average and achieves a 1.29% average speedup over the state-of-the-art, with improvements of up to 22%, narrowing the gap to an ideal predictor to just 1.5%.
The final contribution addresses instruction fusion. Prior work on non-consecutive fusion relies on a Tournament branch predictor with long histories, a design that we found to be inappropriate for this task. To address these shortcomings, we present FLIP, a new fusion predictor that applies the same context-sensitive principle as PHAST. FLIP identifies profitable non-consecutive fusion opportunities among loads and stores and introduces a set of Non-consecutive Fusion Optimizations (NFOs) to relax unnecessary constraints and mitigate harmful cases. Experiments show that FLIP improves performance by 2.44% on SPEC CPU 2017 and 2.94% on MiBench, while reducing MPKI by 83%
Radix Actinidiae chinensis inhibits neovascularization in colorectal cancer and its mechanism
No full textObjective. Colorectal cancer is one of the
most common cancers worldwide, and its angiogenesis
is a key factor in tumor growth and metastasis. Radix
Actinidiae chinensis is considered to have antitumor
activity in traditional Chinese medicine, but its effect on
neovascularization in colorectal cancer has not been
clarified. Herein, we aimed to evaluate the effect of
different concentrations of Radix Actinidiae chinensis on
the neovascularization of colorectal cancer and explore
its possible mechanisms.
Method. A mouse model of colorectal cancer was
established, and mice were randomly divided into
control, low-, and high-concentration groups. Then the
mice in the experimental group were treated with Radix
Actinidiae chinensis, and its effects on neovascularization
and tumor growth were evaluated by tumor growth curve
tracking, immunohistochemical analysis, vessel density
assessment, RT-qPCR, and protein immunoblotting to
explore the underlying mechanisms.
Results. It was shown that tumor tissues in the high
concentration group exhibited significantly slower
growth in both mass and volume compared with the low
concentration and control groups. Immunohistochemical
staining revealed a reduction in the expression of the
vascular endothelial marker CD31 in the Radix
Actinidiae chinensis treatment group. Moreover, the
protein expression levels of vascular markers in tumor
tissues showed a slight decrease in the low-concentration
group and a marked reduction in the high-concentration
group. These findings suggest that angiogenesis in
the tumor microenvironment was inhibited in a
concentration-dependent manner, with protein
expression levels closely mirroring gene expression
patterns.
Conclusion. The study found that Radix Actinidiae
chinensis inhibits neovascularization in a dose
dependent manner in a mouse model of colorectal
cancer. These results provide experimental support for
its potential use as a therapeutic agent against colorectal
cancer, suggesting that it may suppress tumor growth
and metastasis by inhibiting angiogenesis
Trophoblast-derived proteins and their effects on the pathogenesis of preeclampsia
No full textTrophoblast cells are crucial structural units of the placenta, responsible for maintaining its integrity and function. These cells synthesize and secrete specific proteins that play essential roles in placental vascularization, maternal and fetal immune tolerance, and other critical processes. An abnormal level of trophoblast-derived secreted proteins has been closely linked to pregnancy-related diseases. Preeclampsia, a severe complication of pregnancy characterized by de-novo development of hypertension and proteinuria after 20 weeks of gestation, poses significant health risks to both the mother and fetus. This article reviews several key trophoblast-derived proteins that are widely recognized for their roles in preeclampsia. The specific mechanisms of action and interconnections among these proteins in preeclampsia are discussed, along with novel insights into the underlying pathological mechanisms of this disease
Lysimachia christinae Hance aqueous extract ameliorates renal injury in kidney stone rats and calcium oxalate crystal-induced oxidative stress in HK-2 cells via inhibiting the PI3K/Akt/mTOR pathway
No full textObjectives. Kidney stones are a frequent
urinary system disorder. Lysimachia christinae Hance is
an accepted herb in traditional Chinese medicine for
treating kidney stones. However, the effects and
mechanisms of Lysimachia christinae Hance aqueous
extract (LCH) are yet to be elucidated.
Methods. The function of the aqueous extract of
LCH was assessed using kidney stone rat models
induced by 1% ethylene glycol+2% NH4Cl.
Additionally, an in vitro model of human renal tubular
epithelial cells (HK-2) treated with calcium oxalate was
used.
Results. Resultantly, the treatment of aqueous extract
of LCH at different concentrations or LCH+LY294002
(PI3K-specific inhibitor) showed significant improve
ment in inorganic ions and renal pathological injury in
nephrolithiasis rats. Besides, consistent with the in vivo
assay, LCH-containing serum increased cell viability and
inhibited oxidative stress and deposition of Ca2+ in HK
2 cells, while the influences of LCH-containing serum
were attenuated. Mechanistically, the aqueous extract of
LCH and LCH-containing serum also promoted Nrf2
and HO-1 levels and inhibited the phosphorylated
expression levels of PI3K, AKT, and mTOR.
Conclusion. This study shows that LCH ameliorates
the kidney damage in kidney stone rats and HK-2 cells.
The mechanism of LCH in treating kidney stones is
related to the activation of the Nrf2/HO-1 axis and the
inhibition of the PI3K/Akt/mTOR pathway
Determinantes de estilo de vida y cardiometabólicos en el riesgo cardiovascular residual de pacientes anticoagulados con fibrilación auricular
No full textTesis por compendio de publicacionesIntroducción: La fibrilación auricular (FA) es la arritmia sostenida más prevalente y constituye uno de los principales problemas de salud cardiovascular a nivel global. A pesar de que la anticoagulación oral (ACO) reduce de manera significativa la incidencia de ictus isquémico y embolias sistémicas, una proporción considerable de pacientes continúa experimentando eventos adversos. Esto pone de manifiesto las limitaciones de las herramientas tradicionales de estratificación del riesgo, como CHA2DS2-VASc, CHA2DS2-VA y HAS-BLED. Además, la FA se asocia frecuentemente con múltiples comorbilidades, lo que incrementa la heterogeneidad clínica y el riesgo de resultados cardiovasculares desfavorables. Esta tesis, presentada como un compendio de artículos originales, aborda el riesgo residual tromboembólico y cardiovascular en pacientes con FA tratados con ACO. Objetivos: Los objetivos de la tesis son los siguientes: (i) evaluar la adherencia a un estilo de vida saludable como predictor de eventos tromboembólicos y cardiovasculares, (ii) analizar el índice triglicérido-glucosa (TyG) como marcador de resistencia a la insulina y predictor de eventos cardiovasculares adversos, y (iii) estudiar el colesterol remanente (CR) y su discordancia con el colesterol unido a lipoproteínas de baja densidad (LDL-C) como indicador de riesgo tromboembólico y cardiovascular residual. Métodos: La investigación se desarrolló dentro de la cohorte Murcia Atrial Fibrillation Project III (MAFP-III), un registro prospectivo y observacional que incluyó pacientes con diagnóstico reciente de FA y candidatos a tratamiento anticoagulante oral. Los participantes fueron reclutados en dos hospitales de referencia de la Región de Murcia, con criterios de inclusión amplios para garantizar la representatividad de la muestra. La metodología abarcó tres áreas clave: (i) Estilo de vida saludable: se evaluaron factores como presión arterial, índice de masa corporal, glucosa, consumo de tabaco y alcohol. (ii) Índice TyG: calculado como Ln [triglicéridos en ayunas x glucosa en ayunas / 2], utilizado como marcador de resistencia a la insulina. (iii) Discordancia entre LDL-C y CR: el CR se estimó indirectamente con la fórmula colesterol total - colesterol HDL - LDL-C. El seguimiento fue prospectivo y se extendió por dos años, durante los cuales se documentaron eventos tromboembólicos y cardiovasculares, como ictus isquémico, accidente isquémico transitorio, embolismo sistémico, infarto agudo de miocardio y muerte cardiovascular, entre otros. Resultados: Los hallazgos clave fueron los siguientes: (i) Adherencia a un estilo de vida saludable: una alta adherencia a comportamientos saludables al momento del diagnóstico de FA se asoció independientemente con un menor riesgo de eventos tromboembólicos y cardiovasculares. No obstante, menos del 30% de los pacientes alcanzaron una alta adherencia, lo que subraya la necesidad de intervenciones preventivas secundarias; (ii) Índice TyG: valores elevados del índice TyG, indicador de resistencia a la insulina, se asociaron de manera independiente con un mayor riesgo cardiovascular en pacientes con FA tratados con ACO. Este índice se presenta como un biomarcador sencillo, económico y eficaz para una estratificación del riesgo más precisa; y (iii) Colesterol remanente y LDL-C: niveles elevados de CR estuvieron asociados con un mayor riesgo de eventos tromboembólicos, particularmente en pacientes con discordancia entre LDL-C bajo y CR elevado. Este resultado resalta las limitaciones del LDL-C como único marcador lipídico, destacando el papel complementario del CR en la evaluación del riesgo. Conclusiones: Los resultados de esta tesis proporcionan nuevas perspectivas sobre la estratificación de riesgo en pacientes con FA tratados con ACO. Incorporar variables como la adherencia a un estilo de vida saludable, el índice TyG y el CR en los marcos de evaluación de riesgo podría mejorar la identificación de pacientes de alto riesgo de eventos cardiovasculares. Esto permitiría optimizar las estrategias preventivas y apoyar una gestión más personalizada y eficaz de la FA.Introduction: Atrial fibrillation (AF) is the most prevalent sustained arrhythmia and one of the leading cardiovascular health issues globally. Although oral anticoagulation (OAC) significantly reduces the incidence of ischaemic stroke and systemic embolism, a considerable proportion of patients continue to experience adverse events. This highlights the limitations of traditional risk stratification tools, such as CHA2DS2-VASc, CHA2DS2-VA, and HAS-BLED. Furthermore, AF is frequently associated with multiple comorbidities, which increases clinical heterogeneity and the risk of adverse cardiovascular outcomes. This thesis, presented as a compendium of original articles, investigates residual thromboembolic and cardiovascular risk in patients with AF treated with OAC. Objectives: The objectives of the thesis are as follows: (i) to evaluate adherence to a healthy lifestyle as a predictor of thromboembolic and cardiovascular events, (ii) to analyse the triglyceride-glucose (TyG) index as a marker of insulin resistance and predictor of adverse cardiovascular events, and (iii) to study remnant cholesterol (RC) and its discordance with low-density lipoprotein cholesterol (LDL-C) as an indicator of residual thromboembolic and cardiovascular risk .Methods: The research was conducted within the Murcia Atrial Fibrillation Project III (MAFP-III) cohort, a prospective and observational registry that included patients with a recent diagnosis of AF and candidates for oral anticoagulation therapy. Participants were recruited from two reference hospitals in the Region of Murcia, with broad inclusion criteria to ensure sample representativeness. The methodology encompassed three key areas: (i) Healthy lifestyle: factors such as blood pressure, body mass index, glucose, tobacco use, and alcohol consumption were assessed. (ii) TyG index: calculated as Ln [fasting triglycerides x fasting glucose / 2], used as a marker of insulin resistance. (iii) Discordance between LDL-C and RC: RC was estimated indirectly using the formula total cholesterol - HDL cholesterol - LDL-C. The follow-up was prospective, lasting two years, during which thromboembolic and cardiovascular events were documented, such as ischaemic stroke, transient ischaemic attack, systemic embolism, acute myocardial infarction, and cardiovascular death, among others. Results: The key findings were as follows: (i) Adherence to a healthy lifestyle: high adherence to healthy behaviours at the time of AF diagnosis was independently associated with a lower risk of thromboembolic and cardiovascular events. However, fewer than 30% of patients achieved high adherence, highlighting the need for secondary preventive interventions; (ii) TyG index: elevated TyG index values, reflecting insulin resistance, were independently associated with a higher cardiovascular risk in AF patients treated with OAC. This index emerges as a simple, cost-effective biomarker for more precise risk stratification; and (iii) Remnant cholesterol and LDL-C: elevated RC levels were independently associated with a higher risk of thromboembolic events, particularly in patients with discordance between low LDL-C and high RC. This finding underscores the limitations of LDL-C as a sole lipid marker and highlights the complementary role of RC in risk assessment. Conclusions: The findings of this thesis provide new insights into risk stratification in AF patients treated with OAC. Incorporating variables such as adherence to a healthy lifestyle, the TyG index, and RC into risk assessment frameworks could improve the identification of high-risk patients for cardiovascular events. This would allow for the optimisation of preventive strategies, supporting a more personalised and effective management of AF
¿Quién pudo ser Paloma? Fichas de trabajo
No full textEnlace a libro del docente: http://hdl.handle.net/10201/219701Estos materiales educativos se diseñan en el marco del proyecto de investigación "Iniciativa Meletea" financiado por la Fundación Séneca (más info en: https://meletea.um.es/). Se trata de una intervención orientada al uso de metodologías activas y estrategias científico-creativas para una enseñanza interdisciplinar, activa y competencial en Educación Secundaria. Esta intervención se articula bajo el enfoque aprendizaje basado en proyectos, a partir de la pregunta «¿Quién pudo ser Paloma?» y planteando al alumnado una serie de cuestiones abiertas (¿cuándo vivió?, ¿cómo era?, ¿cómo era su entorno?, ¿qué comía?, ¿cómo se comunicaba?), que requieren movilizar conocimientos de Geografía e Historia, Biología y Geología, Matemáticas, Inglés, Expresión Artística y Música. La clave es que el alumnado no recibe respuestas cerradas y unidireccionales, sino que las construye con tareas guiadas y productos parciales, hasta culminar en un producto final cuyo objetivo es la divulgación de los resultados obtenidos. La Prehistoria permite al alumnado iniciarse en la tarea de formular hipótesis, contrastarlas con evidencias científicas y elaborar respuestas argumentadas, en un proceso que reproduce, a escala escolar, algunas fases del trabajo científico
Cómo pensar con nuestros cuerpos: la práctica teórica de Althusser revisitada
No full textEl objetivo de este artículo es revisar la noción de práctica teórica deLouis Althusser para sugerir que esta apunta a un intento muy original de pensar el conocimiento no como algo que tiene que ver con una «conciencia pensante», como sugieren las teorías tradicionales del conocimiento, sino como un proceso y una práctica materiales colectivos. En este sentido, podríamos decir que pensamos con nuestros cuerpos más que con nuestras mentes. Para ello, también revisamos el trabajo de Michel Pêcheux y su conceptualización del conocimiento como un proceso sin sujeto, antes de volver a la petición de Althusser de una nueva práctica materialista de la filosofía.The aim of this article is to revisit Louis Althusser’snotion of theoretical practice in order to suggest that this points to a highly original attempt to think knowledge not as something that has to do with a ‘thinking consciousness’ as traditional theories of knowledge suggest, but as a collective material process and practice. In this sense, we might say that we are thinking with our bodies rather than our minds. To that end we also revisit Michel Pêcheux’s work and his conceptualisation of knowledge as a process without a subject, before returning to Althusser’s call for a new materialist practice of philosophy
Pachymic acid ameliorates polycystic ovary syndrome via inactivating Akt/ERK signaling
No full textObjective. Polycystic ovary syndrome (PCOS) is a common reproductive endocrine disorder that adversely affects women’s health and quality of life. Pachymic acid (PA), a bioactive ingredient from Poria cocos (Schw.) Wolf, has demonstrated protective effects against PCOS in a murine model. This study aims to investigate the underlying mechanism by which PA exerts protective effects against PCOS.
Methods. Female Sprague-Dawley rats were treated with letrozole to induce PCOS. The ovarian granulosa cell line (KGN) was exposed to lipopolysaccharide (LPS) to mimic PCOS in vitro. Hematoxylin-eosin staining and TUNEL assay were used for ovarian histological analysis. The cell counting kit-8 assay was used to assess the viability of KGN cells. Flow cytometry was used for in vitro cell apoptosis analysis. Western blotting revealed molecular protein expression levels in rat ovaries and KGN cells.
Results. PA attenuated LPS-induced lactate dehydrogenase release (p<0.01), reduced the cell apoptosis rate (p<0.001), Bax, and cleaved-caspase3 protein expression (p<0.001), and increased Bcl-2 protein expression (p<0.01) in KGN cells. PA attenuated letrozole-induced increases in testosterone (p<0.01), luteinizing hormone (p<0.01), and estradiol levels (p<0.05) and decreases in progesterone levels (p<0.05) in PCOS rats. PA promoted corpus luteum formation (p<0.001) and reduced the number of cystic follicles and cell apoptosis (p<0.001) in PCOS rats. PA blocked Akt and ERK signaling transduction in PCOS rats and KGN cells (p<0.001).
Conclusion. PA protects against PCOS and attenuates cell apoptosis by inactivating Akt and ERK signaling