Stereotypes in Language & Computational Language Models

Combined knowledge from linguistics, psychology, and natural language processing. In one project they study whether and how generic sentences — which typically express (perhaps stereotypical) generalisations — should be interpreted, and whether corresponding implicit generalisations can be found in computational language models trained by (huge) linguistic corpora. In the other project, the main object of study is the impact of (online) media on stereotypical beliefs people hold.

Understanding and mitigating bias in AI automated systems

“The AI community has been focusing on developing fixes for harmful bias and discrimination, through so-called ‘debiasing algorithms’ that either try to fix data for known or expected biases, or constrain the outcomes of a given predictive model to produce ‘fair’ outcomes. We argue that creating more AI solutions to fix harmful biases in data is not the only solution we should be pursuing. A fundamental question we are facing as researchers and practitioners, is not how to fix harmful bias in AI with new algorithms, but rather; if we should be designing and deploying such potentially biased systems in the first place”.

Building Cultural AI

“Biases in data can be both explicit and implicit. A simple two-word phrase can carry strong contestations, and entire research fields, such as post-colonial studies, are devoted to them. However, these sometimes subtle (and sometimes not so subtle) differences in voice are as yet not often found in the results of automatic analyses or datasets created using automated methods. Current AI technologies and data representations often reflect the popular or majority vote. This is an inherent artefact of the frequentist bias of many statistical analysis methods resulting in simplified representations of the world in which diverse perspectives are underrepresented. In this lecture, I will discuss how the Cultural AI Lab is working towards mitigating this.”

Deep Autoencoders

This lecture overviews Deep Autoencoders that has many applications in image denoising, classification, generation and in object pose estimation. It covers the following topics in detail: unsupervised learning, autoencoder principles, deep autoencoder types (Sparse/variational/ Convolutional/Adversarial Autoencoders) and their applications.

Should you require access to the resource, please contact the author directly.

Artificial Neural Networks. Perceptron

This lecture will cover the basic concepts of Artificial Neural Networks (ANNs): Biological neural models, Perceptron, Activation functions, Loss types, Steepest Gradient Descent, On-line Perceptron training, Batch Perceptron training.

Should you require access to the resource, please contact the author directly.

Neural Image Compression

This lecture overviews Neural Image Compression that has many applications in image storage and communications. It covers the following topics in detail:  Image Compression Types, Image Compression Evaluation, Transform Image Compression, Neural Predictive Image Coding, Neural Image Autoencoding, CNN-Transformer Image Compression, RNN Image Compression, Variable Rate RNN Image Compression.

Should you require access to the resource, please contact the author directly.