Tag: Adaptive Neural Networks
Browse our exclusive articles!
Llama-3.1-8B Uses Base-10 Addition for Cyclic Reasoning
Discover how Llama-3.1-8B language model uses base-10 addition and Fourier features to reason about cyclic concepts like months.
Understanding Representation in Large Language Models
Explore how large language models use representation vs. memorization to process language and their cognitive implications.
Quantization Trap in Multi-Hop Reasoning: Breaking Scaling Laws
Discover how the quantization trap disrupts linear scaling laws, increasing energy use and reducing accuracy in multi-hop reasoning AI models.
DAPPr: Possibilistic Uncertainty for Reliable Deep Learning
Discover DAPPr, a novel framework improving epistemic uncertainty quantification in deep learning for more reliable predictions on unseen data.
Hyperspherical Forward-Forward: Faster AI Training Method
Discover the Hyperspherical Forward-Forward algorithm, a fast, efficient AI training method improving speed and accuracy over traditional approaches.