A Deeper Look into Neural Networks – What is Missing?
Neural networks have become the foundation of modern artificial intelligence, driving breakthroughs in computer vision, natural language processing, and countless real-world applications. Yet despite their success, today’s models still exhibit surprising limitations: they generalize poorly outside their training distribution, struggle with reasoning and abstraction, and often behave as opaque “black boxes.” This seminar takes a deeper look into what neural networks truly do well—and what is still fundamentally missing. We will explore the gaps between pattern recognition and genuine understanding, discuss challenges such as interpretability, robustness, data efficiency, and alignment with human intent, and highlight emerging ideas that aim to move beyond current architectures. The goal is to give students a clearer, more critical perspective on neural networks: not just how they work, but what remains unsolved, and where the next opportunities for innovation lie.
Date and Time
Location
Hosts
Registration
-
Add Event to Calendar
- Big Data and Artificial Intelligence Research Laboratory
- University of Northern British Columbia
- Prince George, British Columbia
- Canada
Speakers
Fan Jiang
A Deeper Look into Neural Networks – What is Missing?
Neural networks have become the foundation of modern artificial intelligence, driving breakthroughs in computer vision, natural language processing, and countless real-world applications. Yet despite their success, today’s models still exhibit surprising limitations: they generalize poorly outside their training distribution, struggle with reasoning and abstraction, and often behave as opaque “black boxes.” This seminar takes a deeper look into what neural networks truly do well—and what is still fundamentally missing. We will explore the gaps between pattern recognition and genuine understanding, discuss challenges such as interpretability, robustness, data efficiency, and alignment with human intent, and highlight emerging ideas that aim to move beyond current architectures. The goal is to give students a clearer, more critical perspective on neural networks: not just how they work, but what remains unsolved, and where the next opportunities for innovation lie.