technique

adaptive-rank instantiation

techniqueactiveprovisionaladaptive-rank-instantiation-cad271b6·1 events·first seen 15d ago

Aliases: adaptive-rank instantiation

Co-occurring entities

Multimodal Large Language Models CRAM centroid-guided routing Mixture of Experts Multimodal Continual Instruction Tuning

More like this (12)

rank-1 approximation reranking LoRA (Low-Rank Adaptation)likelihood approximation reranker models distributed optimization Preference-Aware Rubric Learning multi-turn agent benchmarks hybrid dense-sparse retrieval stochastic gradient ascent Reference-Augmented Training randomized coordinate descent

Recent events (1)

5arXiv · cs.CL·15d ago·source ↗

CRAM: Centroid-Routing and Adaptive MoE for Multimodal Continual Instruction Tuning

CRAM is a new method for Multimodal Continual Instruction Tuning (MCIT) that addresses the tension between catastrophic forgetting and parameter efficiency in MLLMs. It combines adaptive-rank instantiation to dynamically allocate parameters based on capability gaps, centroid-guided routing to reuse existing expert knowledge, and an orthogonality penalty to confine new updates to task-specific directions. The approach uses a Mixture-of-Experts architecture where task-specific patterns are isolated into independent modules, avoiding both the interference of shared updates and the parameter bloat of fully isolated expansion. Experiments across diverse benchmarks show consistent improvements over existing MCIT methods.

Enterprise Deployment Patterns Agent and Tool Ecosystem Multimodal Large Language Models CRAM centroid-guided routing +4 more