Search

Digitalised recruitment processes typically rely on key information automatically extracted from resumes. The case of educational background information is particularly noisy, considering the ever-growing naming of degrees, thus making its normalisation a decisive aspect for subsequent exploitation of such data. In this work we define the normalisation of education information as its transformation into pairs of level/field-of-study. Towards that purpose, we define and share a new taxonomy for fields of study within the labour context. We develop a simple approach where level of study is identified using expert rules, and field of study is normalised using a combination of rules to cover the most frequent occurrences and classifier predictions to generalise over the less frequent cases. We evaluate the proposed system on a new test set that we also make publicly available. We also investigate the application of education normalisation to a candidate-job matching use case., Los procesos de contratación digitalizados suelen basarse en información clave extraída automáticamente de los currículums. El caso de la trayectoria educativa es especialmente conflictivo, considerando la creciente cantidad de titulaciones, por lo que su normalización es decisiva para la posterior explotación de dichos datos. En este trabajo definimos la normalización de la información educativa como su transformación en una serie de pares nivel-campo de estudio. Para ello, definimos y compartimos una nueva taxonomía de campos de estudio en el contexto laboral. Desarrollamos un sistema sencillo en el que el nivel de estudios se identifica mediante reglas expertas, y el campo de estudios se normaliza utilizando una combinación de reglas para cubrir las ocurrencias más frecuentes y predicciones de clasificadores para generalizar sobre los casos menos frecuentes. Evaluamos el sistema propuesto en un nuevo test set compartido públicamente y probamos su aplicación en un caso de uso de comparación candidato-empleo.

Meta-reinforcement learning has the potential to enable artificial agents to master new skills with improved sample-efficiency by leveraging previous learning experience in tasks that are diverse but share common structure. Our focus is to study the application of such algorithms to task distributions where the function that controls the dynamics of the environment is the main factor of variation. We start by providing an introductory background for related fields, including deep reinforcement learning, variational inference, and meta-learning. Then, we conduct a non-systematic review of the state-of-the-art algorithms for meta-reinforcement learning and perform an empirical investigation of PEARL, a method that combines soft actor-critic with latent task variables. Based on our review, we propose and implement two algorithmic modifications for PEARL: one that aims to improve the meta-training sample complexity by automatically adjusting a critical hyperparameter, and a second one focused on improving the meta-testing asymptotic performance by fine-tuning the policy during adaptation. Using a new multi-task environment suite for simulated robotics continuous control tasks, we compare the original version of PEARL and our proposed modifications, obtaining favourable results. Finally, we ponder our findings and suggest future research directions.