Nonetheless, their performance heavily depends on the specificity of the input text prompts, which requires skillful prompt template engineering. Instead, current approaches to prompt optimization learn the prompts through gradient descent, where the prompts are treated as adjustable parameters. However, these methods tend to lead to overfitting of the base classes seen during training and produce prompts that are no longer understandable by humans.

Pre-trained vision-language models like CLIP have remarkably adapted to various downstream tasks. Nonetheless, their performance heavily depends on the specificity of the input text prompts, which requires skillful prompt template engineering. Instead, current approaches to prompt optimization learn the prompts through gradient descent, where the prompts are treated as adjustable parameters. However, these methods tend to lead to overfitting of the base classes seen during training and produce prompts that are no longer understandable by humans. This paper introduces a simple but interpretable prompt optimizer (IPO), that utilizes large language models (LLMs) to generate textual prompts dynamically. We introduce a Prompt Optimization Prompt that not only guides LLMs in creating effective prompts but also stores past prompts with their performance metrics, providing rich in-context information. Additionally, we incorporate a large multimodal model (LMM) to condition on visual content by generating image descriptions, which enhance the interaction between textual and visual modalities. This allows for thae creation of dataset-specific prompts that improve generalization performance, while maintaining human comprehension. Extensive testing across 11 datasets reveals that IPO not only improves the accuracy of existing gradient-descent-based prompt learning methods but also considerably enhances the interpretability of the generated prompts. By leveraging the strengths of LLMs, our approach ensures that the prompts remain human-understandable, thereby facilitating better transparency and oversight for vision-language models.

This paper focuses on task-agnostic prompt compression for better generalizability and efficiency. Considering the redundancy in natural language, existing approaches compress prompts by removing tokens or lexical units according to their information entropy obtained from a causal language model such as LLaMa-7B. The challenge is that information entropy may be a suboptimal compression metric: (i) it only leverages unidirectional context and may fail to capture all essential information needed for prompt compression; (ii) it is not aligned with the prompt compression objective. To address these issues, we propose a data distillation procedure to derive knowledge from an LLM to compress prompts without losing crucial information, and meantime, introduce an extractive text compression dataset. We formulate prompt compression as a token classification problem to guarantee the faithfulness of the compressed prompt to the original one, and use a Transformer encoder as the base architecture to capture all essential information for prompt compression from the full bidirectional context. Our approach leads to lower latency by explicitly learning the compression objective with smaller models such as XLM-RoBERTa-large and mBERT. We evaluate our method on both in-domain and out-of-domain datasets, including MeetingBank, LongBench, ZeroScrolls, GSM8K, and BBH. Despite its small size, our model shows significant performance gains over strong baselines and demonstrates robust generalization ability across different LLMs. Additionally, our model is 3x-6x faster than existing prompt compression methods, while accelerating the end-to-end latency by 1.6x-2.9x with compression ratios of 2x-5x.

Spectral detection technology, as a non-invasive method for rapid detection of substances, combined with deep learning algorithms, has been widely used in food detection. However, in real scenarios, acquiring and labeling spectral data is an extremely labor-intensive task, which makes it impossible to provide enough high-quality data for training efficient supervised deep learning models. To better leverage limited samples, we apply pre-training & fine-tuning paradigm to the field of spectral detection for the first time and propose a pre-training method of deep bidirectional transformers for spectral classification of Chinese liquors, abbreviated as Spectrum-BERT. Specifically, first, to retain the model's sensitivity to the characteristic peak position and local information of the spectral curve, we innovatively partition the curve into multiple blocks and obtain the embeddings of different blocks, as the feature input for the next calculation. Second, in the pre-training stage, we elaborately design two pre-training tasks, Next Curve Prediction (NCP) and Masked Curve Model (MCM), so that the model can effectively utilize unlabeled samples to capture the potential knowledge of spectral data, breaking the restrictions of the insufficient labeled samples, and improving the applicability and performance of the model in practical scenarios. Finally, we conduct a large number of experiments on the real liquor spectral dataset. In the comparative experiments, the proposed Spectrum-BERT significantly outperforms the baselines in multiple metrics and this advantage is more significant on the imbalanced dataset. Moreover, in the parameter sensitivity experiment, we also analyze the model performance under different parameter settings, to provide a reference for subsequent research.