分享

Abstractive Speech Summarization (SSum) aims to generate human-like text summaries from spoken content. It encounters difficulties in handling long speech input and capturing the intricate cross-modal mapping between long speech inputs and short text summaries. Research on large language models (LLMs) and multimodal information fusion has provided new insights for addressing these challenges. In this paper, we propose an end-to-end SSum model that utilizes Q-Former as a connector for the audio-text modality and employs LLMs to generate text summaries directly from speech features. We adopt a multi-stage training approach that includes LLM based ASR and Text Summarization (TSum) tasks as auxiliary tasks. ASR tasks are used to align feature spaces and enhance the LLM's ability to handle longer speech. Then, we utilize a curriculum learning strategy to facilitate the model's transition from TSum to SSum. Finally, our model achieves competitive performance on the How-2 dataset.

In this paper, we present our champion solution to the Global Artificial Intelligence Technology Innovation Competition Track 1: Medical Imaging Diagnosis Report Generation. We select CPT-BASE as our base model for the text generation task. During the pre-training stage, we delete the mask language modeling task of CPT-BASE and instead reconstruct the vocabulary, adopting a span mask strategy and gradually increasing the number of masking ratios to perform the denoising auto-encoder pre-training task. In the fine-tuning stage, we design iterative retrieval augmentation and noise-aware similarity bucket prompt strategies. The retrieval augmentation constructs a mini-knowledge base, enriching the input information of the model, while the similarity bucket further perceives the noise information within the mini-knowledge base, guiding the model to generate higher-quality diagnostic reports based on the similarity prompts. Surprisingly, our single model has achieved a score of 2.321 on leaderboard A, and the multiple model fusion scores are 2.362 and 2.320 on the A and B leaderboards respectively, securing first place in the rankings.

Product attributes are crucial for e-commerce platforms, supporting applications like search, recommendation, and question answering. The task of Product Attribute and Value Identification (PAVI) involves identifying both attributes and their values from product information. In this paper, we formulate PAVI as a generation task and provide, to the best of our knowledge, the most comprehensive evaluation of PAVI so far. We compare three different attribute-value generation (AVG) strategies based on fine-tuning encoder-decoder models on three datasets. Experiments show that end-to-end AVG approach, which is computationally efficient, outperforms other strategies. However, there are differences depending on model sizes and the underlying language model. The code to reproduce all experiments is available at: https://github.com/kassemsabeh/pavi-avg

This paper describes NAIST's submission to the simultaneous track of the IWSLT 2024 Evaluation Campaign: English-to-{German, Japanese, Chinese} speech-to-text translation and English-to-Japanese speech-to-speech translation. We develop a multilingual end-to-end speech-to-text translation model combining two pre-trained language models, HuBERT and mBART. We trained this model with two decoding policies, Local Agreement (LA) and AlignAtt. The submitted models employ the LA policy because it outperformed the AlignAtt policy in previous models. Our speech-to-speech translation method is a cascade of the above speech-to-text model and an incremental text-to-speech (TTS) module that incorporates a phoneme estimation model, a parallel acoustic model, and a parallel WaveGAN vocoder. We improved our incremental TTS by applying the Transformer architecture with the AlignAtt policy for the estimation model. The results show that our upgraded TTS module contributed to improving the system performance.

In this paper, we present MasonTigers' participation in SemEval-2024 Task 10, a shared task aimed at identifying emotions and understanding the rationale behind their flips within monolingual English and Hindi-English code-mixed dialogues. This task comprises three distinct subtasks - emotion recognition in conversation for Hindi-English code-mixed dialogues, emotion flip reasoning for Hindi-English code-mixed dialogues, and emotion flip reasoning for English dialogues. Our team, MasonTigers, contributed to each subtask, focusing on developing methods for accurate emotion recognition and reasoning. By leveraging our approaches, we attained impressive F1-scores of 0.78 for the first task and 0.79 for both the second and third tasks. This performance not only underscores the effectiveness of our methods across different aspects of the task but also secured us the top rank in the first and third subtasks, and the 2nd rank in the second subtask. Through extensive experimentation and analysis, we provide insights into our system's performance and contributions to each subtask.

In this paper, we introduce a novel approach for large language model merging via black-box multi-objective optimization algorithms. The goal of model merging is to combine multiple models, each excelling in different tasks, into a single model that outperforms any of the individual source models. However, model merging faces two significant challenges: First, existing methods rely heavily on human intuition and customized strategies to tackle multiple tasks. Second, it's difficult to search for the great model merging configuration in limited evaluations. To address these challenges, we propose a multi-objective optimization based model merging method named MM-MO. The proposed method can automatically search merging configurations for multiple tasks with multi-objective optimization algorithms. Moreover, to obtain high-quality model merging configurations within a limited number of evaluation iterations, we have made several improvements to multi-objective Bayesian optimization specifically for model merging scenarios. First, we introduced a weak-to-strong method to improve the acquisition strategy. Second, we employed Fisher information to select configurations, further increasing the chances of discovering superior model merging configurations. Third, we designed a sparsity metric as an additional optimization objective to enhance the model's generalization performance across different tasks. We conducted comprehensive experiments with other mainstream model merging methods, demonstrating that our method consistently outperforms them. Moreover, performance improvements are observed even on the tasks not explicitly targeted as optimization objectives, indicating that our method enhances the overall potential of the model. ...

Synthetic data is increasingly being used to address the lack of labeled images in uncommon domains for deep learning tasks. A prominent example is 2D pose estimation of animals, particularly wild species like zebras, for which collecting real-world data is complex and impractical. However, many approaches still require real images, consistency and style constraints, sophisticated animal models, and/or powerful pre-trained networks to bridge the syn-to-real gap. Moreover, they often assume that the animal can be reliably detected in images or videos, a hypothesis that often does not hold, e.g. in wildlife scenarios or aerial images. To solve this, we use synthetic data generated with a 3D photorealistic simulator to obtain the first synthetic dataset that can be used for both detection and 2D pose estimation of zebras without applying any of the aforementioned bridging strategies. Unlike previous works, we extensively train and benchmark our detection and 2D pose estimation models on multiple real-world and synthetic datasets using both pre-trained and non-pre-trained backbones. These experiments show how the models trained from scratch and only with synthetic data can consistently generalize to real-world images of zebras in both tasks. Moreover, we show it is possible to easily generalize those same models to 2D pose estimation of horses with a minimal amount of real-world images to account for the domain transfer. Code, results, trained models; and the synthetic, training, and validation data, including 104K manually labeled frames, are provided as open-source at https://zebrapose.is.tue.mpg.de/

Large language models (LLMs) have demonstrated remarkable capabilities. Their powerful generative abilities enable flexible responses based on various queries or instructions. Emerging as widely adopted generalists for diverse tasks, LLMs are still vulnerable to backdoors. This paper proposes an editing-based generative backdoor, named MEGen, aiming to create a customized backdoor for NLP tasks with the least side effects. In our approach, we first leverage a language model to insert a trigger selected on fixed metrics into the input, then design a pipeline of model editing to directly embed a backdoor into an LLM. By adjusting a small set of local parameters with a mini-batch of samples, MEGen significantly enhances time efficiency and achieves high robustness. Experimental results indicate that our backdoor attack strategy achieves a high attack success rate on poison data while maintaining the model's performance on clean data. Notably, the backdoored model, when triggered, can freely output pre-set dangerous information while successfully completing downstream tasks. This suggests that future LLM applications could be guided to deliver certain dangerous information, thus altering the LLM's generative style. We believe this approach provides insights for future LLM applications and the execution of backdoor attacks on conversational AI systems.

Contextualized ASR models have been demonstrated to effectively improve the recognition accuracy of uncommon phrases when a predefined phrase list is available. However, these models often struggle with bilingual settings, which are prevalent in code-switching speech recognition. In this study, we make the initial attempt to address this challenge by introducing a Cross-lingual Contextual Biasing(XCB) module. Specifically, we augment a pre-trained ASR model for the dominant language by integrating an auxiliary language biasing module and a supplementary language-specific loss, aimed at enhancing the recognition of phrases in the secondary language. Experimental results conducted on our in-house code-switching dataset have validated the efficacy of our approach, demonstrating significant improvements in the recognition of biasing phrases in the secondary language, even without any additional inference overhead. Additionally, our proposed system exhibits both efficiency and generalization when is applied by the unseen ASRU-2019 test set.

Prompt engineering has demonstrated remarkable success in enhancing the performance of large language models (LLMs) across diverse tasks. However, most existing prompt optimization methods only focus on the task-level performance, overlooking the importance of query-preferred prompts, which leads to suboptimal performances. Additionally, these methods rely heavily on frequent interactions with LLMs to obtain feedback for guiding the optimization process, incurring substantial redundant interaction costs. In this paper, we introduce Query-dependent Prompt Optimization (QPO), which leverages multi-loop offline reinforcement learning to iteratively fine-tune a small pretrained language model to generate optimal prompts tailored to the input queries, thus significantly improving the prompting effect on the large target LLM. We derive insights from offline prompting demonstration data, which already exists in large quantities as a by-product of benchmarking diverse prompts on open-sourced tasks, thereby circumventing the expenses of online interactions. Furthermore, we continuously augment the offline dataset with the generated prompts in each loop, as the prompts from the fine-tuned model are supposed to outperform the source prompts in the original dataset. These iterative loops bootstrap the model towards generating optimal prompts. Experiments on various LLM scales and diverse NLP and math tasks demonstrate the efficacy and cost-efficiency of our method in both zero-shot and few-shot scenarios.

This paper presents and evaluates work on the development of an artificial intelligence (AI) anti-bullying system. The system is designed to identify coordinated bullying attacks via social media and other mechanisms, characterize them and propose remediation and response activities to them. In particular, a large language model (LLM) is used to populate an enhanced expert system-based network model of a bullying attack. This facilitates analysis and remediation activity - such as generating report messages to social media companies - determination. The system is described and the efficacy of the LLM for populating the model is analyzed herein.

For single image defocus deblurring, acquiring well-aligned training pairs (or training triplets), i.e., a defocus blurry image, an all-in-focus sharp image (and a defocus blur map), is an intricate task for the development of deblurring models. Existing image defocus deblurring methods typically rely on training data collected by specialized imaging equipment, presupposing that these pairs or triplets are perfectly aligned. However, in practical scenarios involving the collection of real-world data, direct acquisition of training triplets is infeasible, and training pairs inevitably encounter spatial misalignment issues. In this work, we introduce a reblurring-guided learning framework for single image defocus deblurring, enabling the learning of a deblurring network even with misaligned training pairs. Specifically, we first propose a baseline defocus deblurring network that utilizes spatially varying defocus blur map as degradation prior to enhance the deblurring performance. Then, to effectively learn the baseline defocus deblurring network with misaligned training pairs, our reblurring module ensures spatial consistency between the deblurred image, the reblurred image and the input blurry image by reconstructing spatially variant isotropic blur kernels. Moreover, the spatially variant blur derived from the reblurring module can serve as pseudo supervision for defocus blur map during training, interestingly transforming training pairs into training triplets. Additionally, we have collected a new dataset specifically for single image defocus deblurring (SDD) with typical misalignments, which not only substantiates our proposed method but also serves as a benchmark for future research.