8 papers
GPrune-LLM: Generalization-Aware Structured Pruning for Large Language Models
Xiaoyun Liu, Divya Saxena, Jiannong Cao +3
Structured pruning is widely applied to compress large language models (LLMs), but its performance depends heavily on how neuron importance is estimated. Most existing methods rely…
Overcoming Growth-Induced Forgetting in Task-Agnostic Continual Learning
Yuqing Zhao, Jiannong Cao, Divya Saxena +4
In continual learning (CL), model growth enhances adaptability to new data. However, when model growth is applied improperly, especially in task-agnostic CL, where the entire grown…
Geometry-Consistent 4D Gaussian Splatting for Sparse-Input Dynamic View Synthesis
Yiwei Li, Jiannong Cao, Penghui Ruan +3
Gaussian Splatting has been considered as a novel way for view synthesis of dynamic scenes, which shows great potential in AIoT applications such as digital twins. However, recent…
MGAS: Multi-Granularity Architecture Search for Trade-Off Between Model Effectiveness and Efficiency
Xiaoyun Liu, Divya Saxena, Jiannong Cao +2
Neural architecture search (NAS) has gained significant traction in automating the design of neural networks. To reduce search time, differentiable architecture search (DAS) refram…
Enhancing Motion in Text-to-Video Generation with Decomposed Encoding and Conditioning
Penghui Ruan, Pichao Wang, Divya Saxena +2
Despite advancements in Text-to-Video (T2V) generation, producing videos with realistic motion remains challenging. Current models often yield static or minimally dynamic outputs,…
Inductive Spatial Temporal Prediction Under Data Drift with Informative Graph Neural Network
Jialun Zheng, Divya Saxena, Jiannong Cao +2
Inductive spatial temporal prediction can generalize historical data to predict unseen data, crucial for highly dynamic scenarios (e.g., traffic systems, stock markets). However, e…