Advancing Analytic Class-Incremental Learning through Vision-Language Calibration

Class-incremental learning (CIL) with pre-trained models (PTMs) faces a critical trade-off between efficient adaptation and long-term stability. While analytic learning enables rapid, recursive closed-form updates, its efficacy is often compromised by accumulated errors and feature incompatibility. In this paper, we first conduct a systematic study to dissect the failure modes of PTM-based analytic CIL, identifying representation rigidity as the primary bottleneck. Motivated by this insight, we propose VILA, a novel dual-branch framework that advances analytic CIL via a two-level vision-language calibration strategy. Specifically, we coherently fuse plastic, task-adapted features with a frozen, universal visual anchor at the feature level through geometric calibration, and leverage cross-modal semantic priors at the decision level to rectify prediction bias. This confluence maintains analytic-learning's extreme efficiency while overcoming its inherent brittleness. Extensive experiments across eight benchmarks demonstrate that VILA consistently yields superior performance, particularly in fine-grained and long-sequence scenarios. Our framework harmonizes high-fidelity prediction with the simplicity of analytic learning. Our code is available at https://github.com/byzhaoAI/VILA.

Artificial intelligence (AI) systems have a major memory problem: when they learn a new skill, they often forget what they already knew. To prevent this, standard methods force the AI to repeatedly review past information. This process is very slow and requires massive computing power. We developed a new approach called VILA to solve this. Instead of constantly retraining the whole system, VILA uses a two-part memory. One part focuses deeply on the very first task it sees. The other part acts as a fixed "general knowledge" base that never changes. When learning new things, VILA uses a fast mathematical formula to directly link new concepts to its general knowledge. It also uses basic reasoning rules to clearly separate different concepts, completely skipping the slow retraining process. As a result, VILA can continuously learn new skills in just minutes instead of hours, while keeping top-level accuracy. This breakthrough makes it possible to run highly adaptable AI on normal devices without relying on massive data centers, bringing us closer to smart systems that can efficiently learn on the fly.