Experimentation and research in basic AI and applied to different areas
Description
AI, or artificial intelligence, covers a significant number of techniques and is of interest in many domains and topics. In this manner, it covers machine learning methods, going through computer vision, natural language processing or data mining. These techniques can be applied in many domains: healthcare, finances, manufacturing, transportation, etc.
Within this great variety, the present line of work focuses more specifically on investigating machine learning techniques, computer vision, and their combination to be applied in different fields (e.g., industrial, transportation, etc.), where their combination gives rise to promising results in the automation and improvement of their processes. In addition, we should also highlight the use of pattern recognition techniques in both machine learning and computer vision for object detection, environment recognition, and scene analysis that support decision making. Thus, PRAIA has acquired knowledge and conducts research using different techniques in the fields mentioned above. These techniques range from:
- Data preprocessing techniques, including normalizations, typifications, missing value management, outlier management, imperfect supervision, principal component, independent component extraction, etc.
- Traditional feature extraction techniques based on a priori knowledge, including representation spaces for geometric, statistical, or decision-theoretic based models, metric and pseudo-metric spaces, and encodings oriented to syntactic and structural models.
- Feature identification techniques using ensembles. For tasks where few samples and a large number of variables (features) are available, the techniques traditionally used in the field of shape recognition for feature selection are not directly applicable due to the low statistical significance of the results obtained since the relationship between samples and variables is just the inverse of the desirable one. It is possible to use a technique known as "ensembles" for feature selection to deal with this problem, which appears in many applications.
- Techniques for selecting and combining data attributes or features (horizontal), including supervised and unsupervised, discriminative and non-discriminative, formal and knowledge-based, wrapper or parametric analysis, etc.
- Techniques for selecting, extracting, and combining data instances (vertical), including editing and condensing methods, supervised and unsupervised, discriminative and non-discriminative methods.
- Statistical and geometric recognition techniques, including neighbourhood models (e.g., KNN), kernel-based and margin-based discriminative models (e.g., SVM, RBF...), connectionist models (e.g., Neural Networks, Deep Learning...), decision tree-based (Random Forest, Gradient Boosted Trees), etc.
- Structural and Syntactic Recognition techniques, including models based on stochastic or weighted automata and transducers, grammatical inference, Markov models, discriminative learning, etc.