Papers by Jingfan Zhang
LECO: Improving Early Exiting via Learned Exits and Comparison-based Exiting Mechanism (2023.acl-srw)
Copied to clipboard
| Challenge: | Recent work on dynamic early exiting has neglected the intermediate exits’ architectural designs. |
| Approach: | They propose a framework for learning exits and COmparison-based early exiting to improve PTMs’ early exit performance. |
| Outcome: | The proposed framework achieves the SOTA performance on multi-exit BERT training and dynamic early exiting on pre-trained models. |
MiLoRA: Efficient Mixture of Low-Rank Adaptation for Large Language Models Fine-tuning (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Low-rank adaptation and its mixture-of-experts (MOE) methods are highly effective but introduce significant latency in multi-tenant settings due to the LoRA modules and MOE routers added to multiple linear modules. |
| Approach: | They propose a low-rank adaptation variant that considers each LoRA module as an expert and employs a prompt-aware routing mechanism. |
| Outcome: | Extensive analysis on commonsense reasoning tasks and math reasoning tasks show that MiLoRA outperforms strong PEFT baselines with comparable tunable parameter budgets. |
NAG-NER: a Unified Non-Autoregressive Generation Framework for Various NER Tasks (2023.acl-industry)
Copied to clipboard
| Challenge: | Existing models for general NER tasks require entities to be generated in a predefined order, causing error propagation and inefficient decoding. |
| Approach: | They propose a non-autoregressive generation framework for general NER tasks that generates entities as a set instead of a sequence, avoiding error propagation and inefficient decoding. |
| Outcome: | The proposed model outperforms state-of-the-art models on three benchmark NER datasets and two of our proprietary NER tasks. |
Activation Steering Decoding: Mitigating Hallucination in Large Vision-Language Models through Bidirectional Hidden State Intervention (2025.acl-long)
Copied to clipboard
| Challenge: | Large Vision Language Models (LVLMs) suffer from hallucination where generated textual descriptions fail to align accurately with visual semantics. |
| Approach: | They propose a training-free approach that mitigates hallucination through targeted intervention in the model’s intermediate activations by identifying directional patterns of hallucinism in the activation space using a small calibration set. |
| Outcome: | The proposed approach reduces hallucination across multiple benchmarks while maintaining performance on general visual understanding tasks. |
AutoGUI: Scaling GUI Grounding with Automatic Functionality Annotations from LLMs (2025.acl-long)
Copied to clipboard
| Challenge: | Existing datasets for UI-VLMs contain large-scale context-free element annotations or contextualized functional descriptions for elements at a small scale. |
| Approach: | They propose an auto-annotation pipeline that generates massive UI element functionality annotations based on UI content changes induced by interacting with the elements. |
| Outcome: | The proposed pipeline improves accuracy and scales well with human evaluation of a high-quality AutoGUI-704k dataset. |
SCOPE: Preserving Modality-Specific Cues to Mitigate Modality Laziness in Multimodal Learning (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing approaches to learning multimodal representations emphasize shared semantics and overlook modality-specific cues. |
| Approach: | They propose a framework for learning complete multimodal representations using shared and practical cues. |
| Outcome: | SCOPE outperforms SOTA benchmarks on four datasets and achieves 27.10% accuracy improvement. |