Papers by Zhanming Jie
ReFT: Reasoning with Reinforced Fine-Tuning (2024.acl-long)
Copied to clipboard
| Challenge: | Existing approaches to improve the generalization of large language models are using Supervised Fine-Tuning (SFT) this approach does not show sufficient generalization ability because it only relies on the given CoT data. |
| Approach: | They propose to use Chain-of-Thought annotations to train Large Language Models using supervised fine-tuning to improve generalization. |
| Outcome: | The proposed approach outperforms SFT on GSM8K, MathQA, and SVAMP datasets and shows a superior generalization ability. |
Dependency-based Hybrid Trees for Semantic Parsing (D18-1)
Copied to clipboard
| Challenge: | Existing models for semantic parsing focus on structure-based models, but none deal with dependency information. |
| Approach: | They propose a dependency-based hybrid tree model which converts natural language utterances into machine interpretable meaning representations. |
| Outcome: | The proposed model achieves state-of-the-art performance across eight languages and is highly tractable inferenceable. |
Leveraging Training Data in Few-Shot Prompting for Numerical Reasoning (2023.findings-acl)
Copied to clipboard
| Challenge: | Chain-of-thought (CoT) prompts can be challenging to design for arithmetic word problem solving. |
| Approach: | They propose to use training data to replace CoT with programs as the reasoning step . their results show that leveraging training data can improve generalization ability of prompts . |
| Outcome: | The proposed methods improve the generalization ability of prompts and the performance of fine-tuned smaller models in arithmetic word problem solving. |
Better Feature Integration for Named Entity Recognition (2021.naacl-main)
Copied to clipboard
| Challenge: | Existing approaches to named entity recognition (NER) focus on stacking the LSTM and graph neural networks (GCNs) however, the exact interaction mechanism between the two types of features is not clear and the performance gain is not significant. |
| Approach: | They propose a model that incorporates both types of features with a Synergized-LSTM which captures how the two types of feature interact. |
| Outcome: | The proposed model achieves better performance than previous approaches while requiring fewer parameters. |
Non-Autoregressive Machine Translation as Constrained HMM (2024.findings-acl)
Copied to clipboard
| Challenge: | Autoregressive (AR) models have some drawbacks due to slow inference speed and label bias due to local normalization. |
| Approach: | They propose to use a left-to-right Hidden Markov Model (HMM) to control label bias in non-autoregressive translation (NAT) They propose a bi-directional HMM, which can regularize each other's biases via shared parameters. |
| Outcome: | The proposed models can achieve comparable performance to autoregressive Transformers using various decoding methods. |
Dependency-Guided LSTM-CRF for Named Entity Recognition (D19-1)
Copied to clipboard
| Challenge: | Named entity recognition (NER) is one of the most important and fundamental tasks in natural language processing (NLP). |
| Approach: | They propose a dependency-guided model to encode dependency trees and capture their properties for named entity recognition. |
| Outcome: | The proposed model improves named entity recognition performance on standard datasets. |
Better Modeling of Incomplete Annotations for Named Entity Recognition (N19-1)
Copied to clipboard
| Challenge: | Existing approaches to named entity recognition (NER) assume that the training data is fully annotated with named entity information. |
| Approach: | They propose a supervised setup for named entity recognition where annotated data is assumed to be available during training. |
| Outcome: | The proposed approach is able to recognize named entities with incomplete annotations. |
Learning to Reason Deductively: Math Word Problem Solving as Complex Relation Extraction (2022.acl-long)
Copied to clipboard
| Challenge: | Existing approaches to solve math word problems do not provide explanations for generated expressions. |
| Approach: | They propose a deductive approach that presents explainable deductive reasoning steps to iteratively construct target expressions. |
| Outcome: | The proposed model significantly outperforms existing strong baselines on four benchmark datasets. |
To be Closer: Learning to Link up Aspects with Opinions (2021.emnlp-main)
Copied to clipboard
| Challenge: | Dependency parsers are not designed for capturing interaction between opinion words and aspect words. |
| Approach: | They propose to learn an aspect-centric tree structure to shorten distance between aspects and opinion words. |
| Outcome: | The proposed model outperforms baselines on five aspect-based sentiment datasets. |
ENT-DESC: Entity Description Generation by Exploring Knowledge Graph (2020.emnlp-main)
Copied to clipboard
| Challenge: | Existing models for knowledge-to-text generation use RDF triples or key-value pairs to generate a natural language description. |
| Approach: | They propose a large-scale dataset to facilitate the study of KG-to-text . they propose MGCN model architecture that incorporates aggregation methods to extract the rich graph information. |
| Outcome: | The proposed model can represent the original graph information more comprehensively and integrates multiple aggregation methods to extract the rich graph information. |