[KongchangAI]
PaperEvent-Anchored and Task-Adaptive Representation

EASTAR

EASTAR is an event-anchored, task-adaptive framework for learning urban region representations. It uses event anchors to decompose aggregated mobility representations into multiple purpose-related semantic channels, and employs a task-adaptive prompt router to adapt to downstream tasks.

Timeline (last 90 days)

Nov 13

EASTAR is an event-anchored and task-adaptive urban region representation learning framework.

Unverified50%
Nov 13

EASTAR uses offline activity events as sparse semantic anchors and decomposes aggregated mobility representations into multiple purpose-related semantic channels.

Unverified50%
Nov 13

EASTAR introduces free channels to capture latent mobility patterns not covered by event data, and sets up a global channel to preserve overall spatial structure.

Unverified50%
Nov 13

EASTAR designs a task-adaptive prompt router that initializes task-channel correlations from LLM-derived priors and refines routing weights through downstream supervision.

Unverified50%
Nov 13

Across four prediction tasks on two real urban datasets, EASTAR consistently outperformed representative baselines, reducing MAE by up to 39.15%.

Unverified50%
Nov 13

Ablation and interpretability analyses verified the effectiveness of event anchoring, semantic channel construction, and task-adaptive routing.

Unverified50%
Nov 13

EASTAR's two-stage design of "LLM prior + supervised fine-tuning" is essentially transferring the adapter or prompt tuning idea from parameter-efficient transfer learning to the urban representation learning setting.

Unverified50%
Nov 13

In the future, EASTAR is expected to extend to more urban analysis scenarios and further explore handling the sparsity of event data and cross-city transfer capability.

Unverified50%

All Facts (8)

Source Articles