Compositional Learning Journal Club
❓ What Are Compositional Problems?
Compositionality refers to the ability to understand and create new combinations using familiar components, functioning much like an algebraic process. While the human brain is naturally adept at learning and applying compositional principles, neural networks (NNs) struggle with this capability. NNs find it difficult to identify and store shared skills across different tasks and to recombine them in a structured, hierarchical way to tackle new problems. This approach is crucial for tasks like:
- 🗣️ Natural Language Understanding: Interpreting and generating language by combining words and phrases in meaningful ways.
- 👁️ Vision and Perception: Generating, recognizing and assembling different elements of a scene, such as objects and their relationships.
- 🧩 Reasoning and Planning: Decomposing complex decisions into a series of logical steps or actions.
In essence, compositional learning seeks to enable AI systems to think in a modular and structured way, much like how humans break down and understand complex concepts.
The inability of Deep Learning to perform compositional learning is one of the main reasons for Deep Learning’s most critical limitations, including the need to feed them tons of data
📝 About Us
We gather to discuss and analyze cutting-edge research papers, share insights, and collaborate on ideas related to compositional learning. Whether you’re a seasoned researcher or just starting out in the field, our goal is to foster a deep understanding of the principles and challenges of compositionality in AI @ RIML Lab
📚 Activities
- 📖 Paper Discussions: Regular meetings where we discuss recent papers on compositional learning.
- 🔍 Research Insights: Share and discuss key findings and ideas from ongoing research.
- 🤝 Collaborative Exploration: Engage in collaborative discussions to explore new directions in compositional AI.
👥 Members
📧 Contact
For more information, reach out to us via Telegram: