3 papers
cs.LG2026
The Optimization Landscape of Learning Compacted Context Models
Thomas Villeneuve, Alex Sandomirsky, Charles O'Neill +2
Many works approach continual learning through the lens of infinite context windows. As an agent puts more observation into context (concretely the KV cache), compacting said conte…
cs.LG2026
Still: Amortized KV Cache Compaction in a Single Forward Pass
Charles O'Neill, Alex Sandomirsky, Harry Partridge +2
The KV cache is the memory bottleneck of long-horizon language model deployment. Practically, a deployable compactor must be lightweight enough to call during inference, expressive…
hep-th2022
Conformal Blocks in Three Dimensions
Jean-François Fortin, Jingping Li, Alex Sandomirsky +1
We derive expressions for conformal blocks involving operators with arbitrary spins in 3-dimensional CFTs. We use previous results on the action of the OPE in the embedding space t…