In-Context Instruction Learning

Seonghyeon, Ye,; Hyeonbin, Hwang,; Yang, Sohee; Yun, Hyeongu; Kim, Yireun; Seo, Minjoon

doi:10.48550/arxiv.2302.14691

Cited by 7 publications

(8 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Within the prompt, we also provide two in-context examples [75] that were composed from selected cases in the dataset. Full examples for all prompts in this work can be found in Appendix B.…”

Section: Summary Generationmentioning

confidence: 99%

See 1 more Smart Citation

Learning to Generate Explainable Stock Predictions using Self-Reflective Large Language Models

Koa,

Ma,

et al. 2024

Proceedings of the ACM Web Conference 2024

View full text Add to dashboard Cite

Explaining stock predictions is generally a difficult task for traditional non-generative deep learning models, where explanations are limited to visualizing the attention weights on important texts. Today, Large Language Models (LLMs) present a solution to this problem, given their known capabilities to generate human-readable explanations for their decision-making process. However, the task of stock prediction remains challenging for LLMs, as it requires the ability to weigh the varying impacts of chaotic social texts on stock prices. The problem gets progressively harder with the introduction of the explanation component, which requires LLMs to explain verbally why certain factors are more important than the others. On the other hand, to fine-tune LLMs for such a task, one would need expert-annotated samples of explanation for every stock movement in the training set, which is expensive and impractical to scale.To tackle these issues, we propose our Summarize-Explain-Predict (SEP) framework, which utilizes a verbal self-reflective agent and Proximal Policy Optimization (PPO) that allow a LLM teach itself how to generate explainable stock predictions, in a fully autonomous manner. The reflective agent learns how to explain past stock movements through a self-reasoning process, while the PPO trainer trains the model to generate the most likely explanations given the input texts at test-time. The training samples for the PPO trainer are also the responses generated during the reflective process, which eliminates the need for human annotators. Using our SEP framework, we fine-tune a specialized LLM that can outperform both traditional deep-learning and LLM methods in prediction accuracy and Matthews correlation coefficient, for the stock classification task. To justify the generalization capability of our framework, we further test it on the portfolio construction task, and demonstrate its effectiveness through various portfolio metrics. Our code can be accessed through https://github.com/koa-fin/sep.

show abstract

Section: Summary Generationmentioning

confidence: 99%

“…(2) Similar to the previous summarization prompt, we select two cases from the dataset and manually compose the response trajectories to use as few-shot exemplars [75]. Additionally, the two example cases chosen have specifically one Positive and one Negative movement label, in order to avoid any majority label bias [80].…”

Section: Explanationmentioning

confidence: 99%

Learning to Generate Explainable Stock Predictions using Self-Reflective Large Language Models

Koa,

Ma,

et al. 2024

Proceedings of the ACM Web Conference 2024

View full text Add to dashboard Cite

show abstract

“…• ChatGPT We utilize the official gpt-3.5-turbo API 6 to perform inference by in-context learning with 3 or 8 demonstrations following Ye et al (2023).…”

Section: Experiments Setupmentioning

confidence: 99%

Response Enhanced Semi-supervised Dialogue Query Generation

Huang,

Wang,

Gao

et al. 2024

AAAI

View full text Add to dashboard Cite

Leveraging vast and continually updated knowledge from the Internet has been considered an important ability for a dialogue system. Therefore, the dialogue query generation task is proposed for generating search queries from dialogue histories, which will be submitted to a search engine for retrieving relevant websites on the Internet. In this regard, previous efforts were devoted to collecting conversations with annotated queries and training a query producer (QP) via standard supervised learning. However, these studies still face the challenges of data scarcity and domain adaptation. To address these issues, in this paper, we propose a semi-supervised learning framework -- SemiDQG, to improve model performance with unlabeled conversations. Based on the observation that the search query is typically related to the topic of dialogue response, we train a response-augmented query producer (RA) to provide rich and effective training signals for QP. We first apply a similarity-based query selection strategy to select high-quality RA-generated pseudo queries, which are used to construct pseudo instances for training QP and RA. Then, we adopt the REINFORCE algorithm to further enhance QP, with RA-provided rewards as fine-grained training signals. Experimental results and in-depth analysis of three benchmarks show the effectiveness of our framework in cross-domain and low-resource scenarios. Particularly, SemiDQG significantly surpasses ChatGPT and competitive baselines. Our code is available at \url{https://github.com/DeepLearnXMU/SemiDQG}.

show abstract

“…Recently, LLMs like ChatGPT or LLaMA have achieved groundbreaking success (Touvron et al, 2023a,b). With the increasing scale of LLMs, novel techniques such as ICL (Ye et al, 2023) and Chain of Thought (CoT) emerged. ICL demonstrates that adding detailed instructions and examples to task prompts can significantly enhance task performance, whether in zero-shot inference or supervised training.…”

Section: Related Workmentioning

confidence: 99%

“…Trained on extensive, multi-domain corpora, LLMs assimilate a broad spectrum of common sense and domain-agnostic knowledge, equipping them with the ability to discern nuanced differences and linguistic subtleties across various domains Dillion et al, 2023;Yang et al, 2023b;Luo et al, 2024a). Moreover, emerging in-context learning (ICL) techniques demonstrate that task-specific performance can be significantly amplified by simply incorporating concise, task-relevant instructions, demonstrations, and examples into the prompts (Ye et al, 2023;. Although initial research has begun to probe the potential of LLMs and associated techniques in ABSA (Fei et al, 2023;Scaria et al, 2023;Varia et al, 2022), empirical investigations explicitly focusing on multi-domain ABSA are still notably scarce.…”

Section: Introductionmentioning

confidence: 99%

Metro System Inundation in Zhengzhou, Henan Province, China

Yang

Zhao

2022

Sustainability

View full text Add to dashboard Cite

In this study, we investigated the flooding accident that occurred on Metro Line 5 in the capital city of Zhengzhou, Henan Province, China. On 20 July 2021, owing to an extreme rainstorm, serious inundation occurred in the Wulongkou parking lot of Zhengzhou Metro Line 5 and its surrounding area. Flooding forced a train to stop during operation, resulting in 14 deaths. Based on our preliminary investigation and analysis of this accident, we designed three main control measures to reduce the occurrence of similar accidents and mitigate the impact of similar accidents in the future, given the increasing number of extreme storm weather events in recent years: (1) to conduct subway flood risk assessments and to establish an early warning system, involving real-time monitoring of meteorological information during subway operation and construction; (2) to improve subway flood control emergency plans and to establish a response mechanism for subway flooding; and (3) to strengthen safety awareness training to ensure the orderly evacuation of people after accidents.

show abstract

In-Context Instruction Learning

Cited by 7 publications

References 20 publications

Learning to Generate Explainable Stock Predictions using Self-Reflective Large Language Models

Learning to Generate Explainable Stock Predictions using Self-Reflective Large Language Models

Response Enhanced Semi-supervised Dialogue Query Generation

Metro System Inundation in Zhengzhou, Henan Province, China

Contact Info

Product

Resources

About