Tutorial on Artificial Intelligence - Pieter Barnard

Tutorial on Artificial Intelligence - Pieter Barnard

• 1.
  Tutorial on Artificial Intelligence PieterBarnard COALESCE 26/03/2026
• 2.
  Artificial Intelligence vsMachine Learning • AI can be described as the “study of agents that receive precepts from the environment and perform actions” … while • ML forms a “subfield of AI that studies the ability to improve performance based on experience. Some AI systems use machine learning methods to achieve competence, but some do not.” S. J. Russell and P. Norvig, Artificial intelligence: a modern approach. Pearson, 2016. Image from: https://www.unite.ai/machine-learning-vs-deep-learning-key-differences/
• 3.
  Sharma, S.K. andWang, X., 2019. Toward massive machine type communications in ultra-dense cellular IoT networks: Current issues and machine learning-assisted solutions. IEEE Communications Surveys & Tutorials, 22(1), pp.426-471.
• 4.
  Timeline of AI:From Rules to Foundation Models Symbolic AI • Expert systems • Rule-based reasoning • Knowledge Engineering 1980s 2000s Statistical ML • SVMs, Bayesian models • Feature Engineering • Probabilistic Learning 2010s Brittle, little to no learning Learning from data
• 5.
  Timeline of AI:From Rules to Foundation Models Symbolic AI • Expert systems • Rule-based reasoning • Knowledge Engineering 1980s 2000s Statistical ML • SVMs, Bayesian models • Feature Engineering • Probabilistic Learning 2010s Deep Learning • DNNs • GPUs • AlexNet 2015s Brittle, little to no learning Learning from data Learned representations Krizhevsky, Alex, Ilya Sutskever, and Geoffrey E. Hinton. "Imagenet classification with deep convolutional neural networks." Advances in neural information processing systems 25 (2012).
• 6.
  Timeline of AI:From Rules to Foundation Models Symbolic AI • Expert systems • Rule-based reasoning • Knowledge Engineering 1980s 2000s Statistical ML • SVMs, Bayesian models • Feature Engineering • Probabilistic Learning 2010s Deep Learning • DNNs • GPUs • AlexNet 2015s More Deep Learning • CNNs, LSTMs, RNNs • End-to-End Systems • Big Data + Compute • eXplainable AI Brittle, little to no learning Learning from data Learned representations High Performance, task-specific 2020s Ribeiro, Marco Tulio, Sameer Singh, and Carlos Guestrin. "" Why should i trust you?" Explaining the predictions of any classifier." Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining. 2016.
• 7.
  Timeline of AI:From Rules to Foundation Models Symbolic AI • Expert systems • Rule-based reasoning • Knowledge Engineering 1980s 2000s Statistical ML • SVMs, Bayesian models • Feature Engineering • Probabilistic Learning 2010s Deep Learning • DNNs, RL • GPUs • AlexNet 2015s More Deep Learning • CNNs, RNNs, DRL • End-to-End Systems • Big Data + Compute • eXplainable AI Brittle, little to no learning Learning from data Learned representations High Performance, task-specific 2020s Foundation Models • GenAI/Transformers • Neuro-Symbolic AI • LLMs (e.g., BERT, GPT) Towards true general-purpose AI Vaswani, Ashish, et al. "Attention is all you need." Advances in neural information processing systems 30 (2017). Present Yi, Kexin, et al. "Neural-symbolic vqa: Disentangling reasoning from vision and language understanding." Advances in neural information processing systems 31 (2018).
• 8.
  Side Note: Transformersvs RNNs 1. Caleb Writes Code, "Transformer Explained", YouTube, accessible at: https://www.youtube.com/watch?v=nZrZOI0oRuw 2. https://towardsdatascience.com/a-complete-guide-to-bert-with-code-9f87602e4a11/
• 9.
  General Trends inAI: From Models to Systems Traditional ML Paradigm • Model-centric view: optimise a single objective (accuracy, loss) • Static datasets (train/test split) • Limited concern for online improvement Modern AI Systems • End-to-end pipelines: • Data collection & preprocessing • Feature Engineering & embeddings (Data Centric AI) • Model training & validation • Explainability layer • Deployment (APIs, edge devices) • Monitoring (drift, anomalies) • Continuous retraining • Integration with real-world systems (networks, sensors, control loops) • Multi-objective optimisation: • Accuracy + latency + robustness + fairness Sculley, David, et al. "Hidden technical debt in machine learning systems." Advances in neural information processing systems 28 (2015). Jakubik, Johannes, et al. "Data-Centric Artificial Intelligence: J. Jakubik et al." Business & Information Systems Engineering 66.4 (2024): 507-515. 1. Rabanser, Stephan, Stephan Günnemann, and Zachary Lipton. "Failing loudly: An empirical study of methods for detecting dataset shift." Advances in Neural Information Processing Systems 32 (2019). 2. Yu, Guo, et al. "Towards fairness-aware multi-objective optimization." Complex & Intelligent Systems 11.1 (2025): 50.
• 10.
  General Trends inAI: From Prediction to Understanding Traditional ML Focus • Optimise predictive accuracy • Black-box models acceptable • Limited interpretability requirements Emerging Needs • Model transparency & interpretability • Regulation (e.g., GDPR “right to explanation”) • Model failures in real-world deployment. • See: https://www.techtarget.com/searchcio/feature/AI- failure-examples-What-real-world-breakdowns-teach- CIOs?utm_source=chatgpt.com • Human-AI collaboration • Debugging model behaviour • Trust in safety-critical systems: • Healthcare • Autonomous systems • Cybersecurity 1. Adadi, Amina, and Mohammed Berrada. "Peeking inside the black-box: a survey on explainable artificial intelligence (XAI)." IEEE access 6 (2018): 52138-52160. 2. Molnar, Christoph. Interpretable machine learning. Lulu. com, 2020. 3. Rudin, Cynthia. "Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead." Nature machine intelligence 1.5 (2019): 206-215. Barnard, Pieter, et al. "Resource reservation in sliced networks: An explainable artificial intelligence (XAI) approach." ICC 2022-IEEE international conference on communications. IEEE, 2022. (Preliminary STRR model) (Final STRR model)
• 11.
  General Trends inAI: From Static to Adaptive systems Traditional ML • Train once, deploy once • Fixed dataset • No feedback loop Modern AI Systems • Continuous learning • Online adaptation • Feedback-driven updates • Approaches • Deep Reinforcement Learning (DRL) • Online learning • Self-supervised adaptation • Applications • Autonomous vehicles • Network optimisation • Robotics • Digital twins 1. Tang, Chen, et al. "Deep reinforcement learning for robotics: A survey of real-world successes." Annual Review of Control, Robotics, and Autonomous Systems 8.1 (2025): 153-188. 2. Govinda, Shruti, Bouziane Brik, and Saad Harous. "A survey on deep reinforcement learning applications in autonomous systems: Applications, open challenges, and future directions." IEEE Transactions on Intelligent Transportation Systems (2025).
• 12.
  Latest Trends https://news.mit.edu/2025/photonic-processor-could- streamline-6g-wireless-signal-processing-0611 https://news.mit.edu/2024/photonic-processor-could- enable-ultrafast-ai-computations-1202 • AImodels are hitting limits of electronic hardware (energy, latency, scale) • Increasing demand from: • Large-scale AI models • Real-time systems (6G, autonomous systems) • Key Breakthrough: Photonic processors perform neural network operations using light • Optical Neural Networks • Up to 90% less energy • ~100× faster than digital hardware • Enables: • Dynamic spectrum management • Ultra-low latency communications
• 13.
  Useful Links forFurther Reading • Causal Structural Models: • https://www.youtube.com/watch?v=zvrcyqcN9Wo&list=PLNSbiazKP Sj-GCffW_eA9hhxcdTSJsg6d • Peters, Jonas, Stefan Bauer, and Niklas Pfister. "Causal models for dynamical systems." Probabilistic and Causal Inference: The Works of Judea Pearl. 2022. 671-690. • https://www.ibm.com/think/topics/causal-inference • Neuro-Symbolic Reasoning • https://mitibmwatsonailab.mit.edu/research/blog/clevrer-the-first- video-dataset-for-neuro-symbolic-reasoning/