Md Zesun Ahmed Mia

RMAAT: Astrocyte-Inspired Memory Compression and Replay for Efficient Long-Context Transformers

The quadratic complexity of self-attention mechanism presents a significant impediment to applying Transformer models to long sequences. This work explores computational principles derived from astrocytes—glial cells critical for biological memory and synaptic modulation—as a complementary approach…