MinMo: A Multimodal Large Language Model for Seamless Voice Interaction

Recent advancements in large language models (LLMs) and multimodal speech-text models have laid the groundwork for seamless voice interactions, enabling real-time, natural, and human-like conversations. Previous models for voice interactions are categorized as native and aligned. Native models integrate speech and text processing in one framework but struggle with issues like differing sequence lengths and insufficient pre-training. Aligned models maintain text LLM capabilities but are often limited by small datasets and a narrow focus on speech tasks. In this work, we introduce MinMo, a Multimodal Large Language Model with approximately 8B parameters for seamless voice interaction. We address the main limitations of prior aligned multimodal models. We train MinMo through multiple stages of speech-to-text alignment, text-to-speech alignment, speech-to-speech alignment, and duplex interaction alignment, on 1.4 million hours of diverse speech data and a broad range of speech tasks. After the multi-stage training, MinMo achieves state-of-the-art performance across various benchmarks for voice comprehension and generation while maintaining the capabilities of text LLMs, and also facilitates full-duplex conversation, that is, simultaneous two-way communication between the user and the system. Moreover, we propose a novel and simple voice decoder that outperforms prior models in voice generation. The enhanced instruction-following capabilities of MinMo supports controlling speech generation based on user instructions, with various nuances including emotions, dialects, and speaking rates, and mimicking specific voices. For MinMo, the speech-to-text latency is approximately 100ms, full-duplex latency is approximately 600ms in theory and 800ms in practice. The MinMo project web page is https://funaudiollm.github.io/minmo, and the code and models will be released soon.

Multimodal Clinical Reasoning through Knowledge-augmented Rationale Generation

Clinical rationales play a pivotal role in accurate disease diagnosis; however, many models predominantly use discriminative methods and overlook the importance of generating supportive rationales. Rationale distillation is a process that transfers knowledge from large language models (LLMs) to smaller language models (SLMs), thereby enhancing the latter's ability to break down complex tasks. Despite its benefits, rationale distillation alone is inadequate for addressing domain knowledge limitations in tasks requiring specialized expertise, such as disease diagnosis. Effectively embedding domain knowledge in SLMs poses a significant challenge. While current LLMs are primarily geared toward processing textual data, multimodal LLMs that incorporate time series data, especially electronic health records (EHRs), are still evolving. To tackle these limitations, we introduce ClinRaGen, an SLM optimized for multimodal rationale generation in disease diagnosis. ClinRaGen incorporates a unique knowledge-augmented attention mechanism to merge domain knowledge with time series EHR data, utilizing a stepwise rationale distillation strategy to produce both textual and time series-based clinical rationales. Our evaluations show that ClinRaGen markedly improves the SLM's capability to interpret multimodal EHR data and generate accurate clinical rationales, supporting more reliable disease diagnosis, advancing LLM applications in healthcare, and narrowing the performance divide between LLMs and SLMs.

EndoFinder: Online Image Retrieval for Explainable Colorectal Polyp Diagnosis

Determining the necessity of resecting malignant polyps during colonoscopy screen is crucial for patient outcomes, yet challenging due to the time-consuming and costly nature of histopathology examination. While deep learning-based classification models have shown promise in achieving optical biopsy with endoscopic images, they often suffer from a lack of explainability. To overcome this limitation, we introduce EndoFinder, a content-based image retrieval framework to find the 'digital twin' polyp in the reference database given a newly detected polyp. The clinical semantics of the new polyp can be inferred referring to the matched ones. EndoFinder pioneers a polyp-aware image encoder that is pre-trained on a large polyp dataset in a self-supervised way, merging masked image modeling with contrastive learning. This results in a generic embedding space ready for different downstream clinical tasks based on image retrieval. We validate the framework on polyp re-identification and optical biopsy tasks, with extensive experiments demonstrating that EndoFinder not only achieves explainable diagnostics but also matches the performance of supervised classification models. EndoFinder's reliance on image retrieval has the potential to support diverse downstream decision-making tasks during real-time colonoscopy procedures.

A Two-Stage Dual-Path Framework for Text Tampering Detection and Recognition

Document tamper detection has always been an important aspect of tamper detection. Before the advent of deep learning, document tamper detection was difficult. We have made some explorations in the field of text tamper detection based on deep learning. Our Ps tamper detection method includes three steps: feature assistance, audit point positioning, and tamper recognition. It involves hierarchical filtering and graded output (tampered/suspected tampered/untampered). By combining artificial tamper data features, we simulate and augment data samples in various scenarios (cropping with noise addition/replacement, single character/space replacement, smearing/splicing, brightness/contrast adjustment, etc.). The auxiliary features include exif/binary stream keyword retrieval/noise, which are used for branch detection based on the results. Audit point positioning uses detection frameworks and controls thresholds for high and low density detection. Tamper recognition employs a dual-path dual-stream recognition network, with RGB and ELA stream feature extraction. After dimensionality reduction through self-correlation percentile pooling, the fused output is processed through vlad, yielding an accuracy of 0.804, recall of 0.659, and precision of 0.913.

Intelligent Parsing: An Automated Parsing Framework for Extracting Design Semantics from E-commerce Creatives

In the industrial e-commerce landscape, creative designs such as banners and posters are ubiquitous. Extracting structured semantic information from creative e-commerce design materials (manuscripts crafted by designers) to obtain design semantics represents a core challenge in the realm of intelligent design. In this paper, we propose a comprehensive automated framework for intelligently parsing creative materials. This framework comprises material recognition, preprocess, smartname, and label layers. The material recognition layer consolidates various detection and recognition interfaces, covering business aspects including detection of auxiliary areas within creative materials and layer-level detection, alongside label identification. Algorithmically, it encompasses a variety of coarse-to-fine methods such as Cascade RCNN, GFL, and other models. The preprocess layer involves filtering creative layers and grading creative materials. The smartname layer achieves intelligent naming for creative materials, while the label layer covers multi-level tagging for creative materials, enabling tagging at different hierarchical levels. Intelligent parsing constitutes a complete parsing framework that significantly aids downstream processes such as intelligent creation, creative optimization, and material library construction. Within the practical business applications at Suning, it markedly enhances the exposure, circulation, and click-through rates of creative materials, expediting the closed-loop production of creative materials and yielding substantial benefits.

Label-dependent and event-guided interpretable disease risk prediction using EHRs

Electronic health records (EHRs) contain patients' heterogeneous data that are collected from medical providers involved in the patient's care, including medical notes, clinical events, laboratory test results, symptoms, and diagnoses. In the field of modern healthcare, predicting whether patients would experience any risks based on their EHRs has emerged as a promising research area, in which artificial intelligence (AI) plays a key role. To make AI models practically applicable, it is required that the prediction results should be both accurate and interpretable. To achieve this goal, this paper proposed a label-dependent and event-guided risk prediction model (LERP) to predict the presence of multiple disease risks by mainly extracting information from unstructured medical notes. Our model is featured in the following aspects. First, we adopt a label-dependent mechanism that gives greater attention to words from medical notes that are semantically similar to the names of risk labels. Secondly, as the clinical events (e.g., treatments and drugs) can also indicate the health status of patients, our model utilizes the information from events and uses them to generate an event-guided representation of medical notes. Thirdly, both label-dependent and event-guided representations are integrated to make a robust prediction, in which the interpretability is enabled by the attention weights over words from medical notes. To demonstrate the applicability of the proposed method, we apply it to the MIMIC-III dataset, which contains real-world EHRs collected from hospitals. Our method is evaluated in both quantitative and qualitative ways.

Label Dependent Attention Model for Disease Risk Prediction Using Multimodal Electronic Health Records

Disease risk prediction has attracted increasing attention in the field of modern healthcare, especially with the latest advances in artificial intelligence (AI). Electronic health records (EHRs), which contain heterogeneous patient information, are widely used in disease risk prediction tasks. One challenge of applying AI models for risk prediction lies in generating interpretable evidence to support the prediction results while retaining the prediction ability. In order to address this problem, we propose the method of jointly embedding words and labels whereby attention modules learn the weights of words from medical notes according to their relevance to the names of risk prediction labels. This approach boosts interpretability by employing an attention mechanism and including the names of prediction tasks in the model. However, its application is only limited to the handling of textual inputs such as medical notes. In this paper, we propose a label dependent attention model LDAM to 1) improve the interpretability by exploiting Clinical-BERT (a biomedical language model pre-trained on a large clinical corpus) to encode biomedically meaningful features and labels jointly; 2) extend the idea of joint embedding to the processing of time-series data, and develop a multi-modal learning framework for integrating heterogeneous information from medical notes and time-series health status indicators. To demonstrate our method, we apply LDAM to the MIMIC-III dataset to predict different disease risks. We evaluate our method both quantitatively and qualitatively. Specifically, the predictive power of LDAM will be shown, and case studies will be carried out to illustrate its interpretability.

Self-Supervised Detection of Contextual Synonyms in a Multi-Class Setting: Phenotype Annotation Use Case

Contextualised word embeddings is a powerful tool to detect contextual synonyms. However, most of the current state-of-the-art (SOTA) deep learning concept extraction methods remain supervised and underexploit the potential of the context. In this paper, we propose a self-supervised pre-training approach which is able to detect contextual synonyms of concepts being training on the data created by shallow matching. We apply our methodology in the sparse multi-class setting (over 15,000 concepts) to extract phenotype information from electronic health records. We further investigate data augmentation techniques to address the problem of the class sparsity. Our approach achieves a new SOTA for the unsupervised phenotype concept annotation on clinical text on F1 and Recall outperforming the previous SOTA with a gain of up to 4.5 and 4.0 absolute points, respectively. After fine-tuning with as little as 20\% of the labelled data, we also outperform BioBERT and ClinicalBERT. The extrinsic evaluation on three ICU benchmarks also shows the benefit of using the phenotypes annotated by our model as features.

Mitigating Backdoor Attacks in Federated Learning

Malicious clients can attack federated learning systems by using malicious data, including backdoor samples, during the training phase. The compromised global model will perform well on the validation dataset designed for the task. However, a small subset of data with backdoor patterns may trigger the model to make a wrong prediction. Previously, there was an arms race. Attackers tried to conceal attacks and defenders tried to detect attacks during the aggregation stage of training on the server-side in a federated learning system. In this work, we propose a new method to mitigate backdoor attacks after the training phase. Specifically, we designed a federated pruning method to remove redundant neurons in the network and then adjust the model's extreme weight values. Experiments conducted on distributed Fashion-MNIST have shown that our method can reduce the average attack success rate from 99.7% to 1.9% with a 5.5% loss of test accuracy on the validation dataset. To minimize the pruning influence on test accuracy, we can fine-tune after pruning, and the attack success rate drops to 6.4%, with only a 1.7% loss of test accuracy.

An Epidemiological Modelling Approach for Covid19 via Data Assimilation

The global pandemic of the 2019-nCov requires the evaluation of policy interventions to mitigate future social and economic costs of quarantine measures worldwide. We propose an epidemiological model for forecasting and policy evaluation which incorporates new data in real-time through variational data assimilation. We analyze and discuss infection rates in China, the US and Italy. In particular, we develop a custom compartmental SIR model fit to variables related to the epidemic in Chinese cities, named SITR model. We compare and discuss model results which conducts updates as new observations become available. A hybrid data assimilation approach is applied to make results robust to initial conditions. We use the model to do inference on infection numbers as well as parameters such as the disease transmissibility rate or the rate of recovery. The parameterisation of the model is parsimonious and extendable, allowing for the incorporation of additional data and parameters of interest. This allows for scalability and the extension of the model to other locations or the adaption of novel data sources.