Hard-Synth: Synthesizing Diverse Hard Samples for ASR using Zero-Shot TTS and LLM

Text-to-speech (TTS) models have been widely adopted to enhance automatic speech recognition (ASR) systems using text-only corpora, thereby reducing the cost of labeling real speech data. Existing research primarily utilizes additional text data and predefined speech styles supported by TTS models. In this paper, we propose Hard-Synth, a novel ASR data augmentation method that leverages large language models (LLMs) and advanced zero-shot TTS. Our approach employs LLMs to generate diverse in-domain text through rewriting, without relying on additional text data. Rather than using predefined speech styles, we introduce a hard prompt selection method with zero-shot TTS to clone speech styles that the ASR model finds challenging to recognize. Experiments demonstrate that Hard-Synth significantly enhances the Conformer model, achieving relative word error rate (WER) reductions of 6.5\%/4.4\% on LibriSpeech dev/test-other subsets. Additionally, we show that Hard-Synth is data-efficient and capable of reducing bias in ASR.

SCIGS: 3D Gaussians Splatting from a Snapshot Compressive Image

Snapshot Compressive Imaging (SCI) offers a possibility for capturing information in high-speed dynamic scenes, requiring efficient reconstruction method to recover scene information. Despite promising results, current deep learning-based and NeRF-based reconstruction methods face challenges: 1) deep learning-based reconstruction methods struggle to maintain 3D structural consistency within scenes, and 2) NeRF-based reconstruction methods still face limitations in handling dynamic scenes. To address these challenges, we propose SCIGS, a variant of 3DGS, and develop a primitive-level transformation network that utilizes camera pose stamps and Gaussian primitive coordinates as embedding vectors. This approach resolves the necessity of camera pose in vanilla 3DGS and enhances multi-view 3D structural consistency in dynamic scenes by utilizing transformed primitives. Additionally, a high-frequency filter is introduced to eliminate the artifacts generated during the transformation. The proposed SCIGS is the first to reconstruct a 3D explicit scene from a single compressed image, extending its application to dynamic 3D scenes. Experiments on both static and dynamic scenes demonstrate that SCIGS not only enhances SCI decoding but also outperforms current state-of-the-art methods in reconstructing dynamic 3D scenes from a single compressed image. The code will be made available upon publication.

SynStitch: a Self-Supervised Learning Network for Ultrasound Image Stitching Using Synthetic Training Pairs and Indirect Supervision

Ultrasound (US) image stitching can expand the field-of-view (FOV) by combining multiple US images from varied probe positions. However, registering US images with only partially overlapping anatomical contents is a challenging task. In this work, we introduce SynStitch, a self-supervised framework designed for 2DUS stitching. SynStitch consists of a synthetic stitching pair generation module (SSPGM) and an image stitching module (ISM). SSPGM utilizes a patch-conditioned ControlNet to generate realistic 2DUS stitching pairs with known affine matrix from a single input image. ISM then utilizes this synthetic paired data to learn 2DUS stitching in a supervised manner. Our framework was evaluated against multiple leading methods on a kidney ultrasound dataset, demonstrating superior 2DUS stitching performance through both qualitative and quantitative analyses. The code will be made public upon acceptance of the paper.

Series-to-Series Diffusion Bridge Model

Diffusion models have risen to prominence in time series forecasting, showcasing their robust capability to model complex data distributions. However, their effectiveness in deterministic predictions is often constrained by instability arising from their inherent stochasticity. In this paper, we revisit time series diffusion models and present a comprehensive framework that encompasses most existing diffusion-based methods. Building on this theoretical foundation, we propose a novel diffusion-based time series forecasting model, the Series-to-Series Diffusion Bridge Model ($\mathrm{S^2DBM}$), which leverages the Brownian Bridge process to reduce randomness in reverse estimations and improves accuracy by incorporating informative priors and conditions derived from historical time series data. Experimental results demonstrate that $\mathrm{S^2DBM}$ delivers superior performance in point-to-point forecasting and competes effectively with other diffusion-based models in probabilistic forecasting.

Audio Is the Achilles' Heel: Red Teaming Audio Large Multimodal Models

Large Multimodal Models (LMMs) have demonstrated the ability to interact with humans under real-world conditions by combining Large Language Models (LLMs) and modality encoders to align multimodal information (visual and auditory) with text. However, such models raise new safety challenges of whether models that are safety-aligned on text also exhibit consistent safeguards for multimodal inputs. Despite recent safety-alignment research on vision LMMs, the safety of audio LMMs remains under-explored. In this work, we comprehensively red team the safety of five advanced audio LMMs under three settings: (i) harmful questions in both audio and text formats, (ii) harmful questions in text format accompanied by distracting non-speech audio, and (iii) speech-specific jailbreaks. Our results under these settings demonstrate that open-source audio LMMs suffer an average attack success rate of 69.14% on harmful audio questions, and exhibit safety vulnerabilities when distracted with non-speech audio noise. Our speech-specific jailbreaks on Gemini-1.5-Pro achieve an attack success rate of 70.67% on the harmful query benchmark. We provide insights on what could cause these reported safety-misalignments. Warning: this paper contains offensive examples.

LMHaze: Intensity-aware Image Dehazing with a Large-scale Multi-intensity Real Haze Dataset

Image dehazing has drawn a significant attention in recent years. Learning-based methods usually require paired hazy and corresponding ground truth (haze-free) images for training. However, it is difficult to collect real-world image pairs, which prevents developments of existing methods. Although several works partially alleviate this issue by using synthetic datasets or small-scale real datasets. The haze intensity distribution bias and scene homogeneity in existing datasets limit the generalization ability of these methods, particularly when encountering images with previously unseen haze intensities. In this work, we present LMHaze, a large-scale, high-quality real-world dataset. LMHaze comprises paired hazy and haze-free images captured in diverse indoor and outdoor environments, spanning multiple scenarios and haze intensities. It contains over 5K high-resolution image pairs, surpassing the size of the biggest existing real-world dehazing dataset by over 25 times. Meanwhile, to better handle images with different haze intensities, we propose a mixture-of-experts model based on Mamba (MoE-Mamba) for dehazing, which dynamically adjusts the model parameters according to the haze intensity. Moreover, with our proposed dataset, we conduct a new large multimodal model (LMM)-based benchmark study to simulate human perception for evaluating dehazed images. Experiments demonstrate that LMHaze dataset improves the dehazing performance in real scenarios and our dehazing method provides better results compared to state-of-the-art methods.

Jigsaw Puzzles: Splitting Harmful Questions to Jailbreak Large Language Models

Large language models (LLMs) have exhibited outstanding performance in engaging with humans and addressing complex questions by leveraging their vast implicit knowledge and robust reasoning capabilities. However, such models are vulnerable to jailbreak attacks, leading to the generation of harmful responses. Despite recent research on single-turn jailbreak strategies to facilitate the development of defence mechanisms, the challenge of revealing vulnerabilities under multi-turn setting remains relatively under-explored. In this work, we propose Jigsaw Puzzles (JSP), a straightforward yet effective multi-turn jailbreak strategy against the advanced LLMs. JSP splits questions into harmless fractions as the input of each turn, and requests LLMs to reconstruct and respond to questions under multi-turn interaction. Our experimental results demonstrate that the proposed JSP jailbreak bypasses original safeguards against explicitly harmful content, achieving an average attack success rate of 93.76% on 189 harmful queries across 5 advanced LLMs (Gemini-1.5-Pro, Llama-3.1-70B, GPT-4, GPT-4o, GPT-4o-mini). Moreover, JSP achieves a state-of-the-art attack success rate of 92% on GPT-4 on the harmful query benchmark, and exhibits strong resistant to defence strategies. Warning: this paper contains offensive examples.

LogLM: From Task-based to Instruction-based Automated Log Analysis

Automatic log analysis is essential for the efficient Operation and Maintenance (O&M) of software systems, providing critical insights into system behaviors. However, existing approaches mostly treat log analysis as training a model to perform an isolated task, using task-specific log-label pairs. These task-based approaches are inflexible in generalizing to complex scenarios, depend on task-specific training data, and cost significantly when deploying multiple models. In this paper, we propose an instruction-based training approach that transforms log-label pairs from multiple tasks and domains into a unified format of instruction-response pairs. Our trained model, LogLM, can follow complex user instructions and generalize better across different tasks, thereby increasing flexibility and reducing the dependence on task-specific training data. By integrating major log analysis tasks into a single model, our approach also relieves model deployment burden. Experimentally, LogLM outperforms existing approaches across five log analysis capabilities, and exhibits strong generalization abilities on complex instructions and unseen tasks.

A Hybrid Model and Learning-Based Force Estimation Framework for Surgical Robots

Haptic feedback to the surgeon during robotic surgery would enable safer and more immersive surgeries but estimating tissue interaction forces at the tips of robotically controlled surgical instruments has proven challenging. Few existing surgical robots can measure interaction forces directly and the additional sensor may limit the life of instruments. We present a hybrid model and learning-based framework for force estimation for the Patient Side Manipulators (PSM) of a da Vinci Research Kit (dVRK). The model-based component identifies the dynamic parameters of the robot and estimates free-space joint torque, while the learning-based component compensates for environmental factors, such as the additional torque caused by trocar interaction between the PSM instrument and the patient's body wall. We evaluate our method in an abdominal phantom and achieve an error in force estimation of under 10% normalized root-mean-squared error. We show that by using a model-based method to perform dynamics identification, we reduce reliance on the training data covering the entire workspace. Although originally developed for the dVRK, the proposed method is a generalizable framework for other compliant surgical robots. The code is available at https://github.com/vu-maple-lab/dvrk_force_estimation.

Context-aware and Style-related Incremental Decoding framework for Discourse-Level Literary Translation

This report outlines our approach for the WMT24 Discourse-Level Literary Translation Task, focusing on the Chinese-English language pair in the Constrained Track. Translating literary texts poses significant challenges due to the nuanced meanings, idiomatic expressions, and intricate narrative structures inherent in such works. To address these challenges, we leveraged the Chinese-Llama2 model, specifically enhanced for this task through a combination of Continual Pre-training (CPT) and Supervised Fine-Tuning (SFT). Our methodology includes a novel Incremental Decoding framework, which ensures that each sentence is translated with consideration of its broader context, maintaining coherence and consistency throughout the text. This approach allows the model to capture long-range dependencies and stylistic elements, producing translations that faithfully preserve the original literary quality. Our experiments demonstrate significant improvements in both sentence-level and document-level BLEU scores, underscoring the effectiveness of our proposed framework in addressing the complexities of document-level literary translation.