In this technical report, we tackle the challenges of training large-scale Mixture of Experts (MoE) models, focusing on overcoming cost inefficiency and resource limitations prevalent in such systems. To address these issues, we present two differently sized MoE large language models (LLMs), namely Ling-Lite and Ling-Plus (referred to as "Bailing" in Chinese, spelled B\v{a}il\'ing in Pinyin). Ling-Lite contains 16.8 billion parameters with 2.75 billion activated parameters, while Ling-Plus boasts 290 billion parameters with 28.8 billion activated parameters. Both models exhibit comparable performance to leading industry benchmarks. This report offers actionable insights to improve the efficiency and accessibility of AI development in resource-constrained settings, promoting more scalable and sustainable technologies. Specifically, to reduce training costs for large-scale MoE models, we propose innovative methods for (1) optimization of model architecture and training processes, (2) refinement of training anomaly handling, and (3) enhancement of model evaluation efficiency. Additionally, leveraging high-quality data generated from knowledge graphs, our models demonstrate superior capabilities in tool use compared to other models. Ultimately, our experimental findings demonstrate that a 300B MoE LLM can be effectively trained on lower-performance devices while achieving comparable performance to models of a similar scale, including dense and MoE models. Compared to high-performance devices, utilizing a lower-specification hardware system during the pre-training phase demonstrates significant cost savings, reducing computing costs by approximately 20%. The models can be accessed at https://huggingface.co/inclusionAI.

This paper aims to address the challenge of hallucinations in Multimodal Large Language Models (MLLMs) particularly for dense image captioning tasks. To tackle the challenge, we identify the current lack of a metric that finely measures the caption quality in concept level. We hereby introduce HalFscore, a novel metric built upon the language graph and is designed to evaluate both the accuracy and completeness of dense captions at a granular level. Additionally, we identify the root cause of hallucination as the model's over-reliance on its language prior. To address this, we propose PerturboLLaVA, which reduces the model's reliance on the language prior by incorporating adversarially perturbed text during training. This method enhances the model's focus on visual inputs, effectively reducing hallucinations and producing accurate, image-grounded descriptions without incurring additional computational overhead. PerturboLLaVA significantly improves the fidelity of generated captions, outperforming existing approaches in handling multimodal hallucinations and achieving improved performance across general multimodal benchmarks.

Collaborative Object Detection (COD) and collaborative perception can integrate data or features from various entities, and improve object detection accuracy compared with individual perception. However, adversarial attacks pose a potential threat to the deep learning COD models, and introduce high output uncertainty. With unknown attack models, it becomes even more challenging to improve COD resiliency and quantify the output uncertainty for highly dynamic perception scenes such as autonomous vehicles. In this study, we propose the Trusted Uncertainty Quantification in Collaborative Perception framework (TUQCP). TUQCP leverages both adversarial training and uncertainty quantification techniques to enhance the adversarial robustness of existing COD models. More specifically, TUQCP first adds perturbations to the shared information of randomly selected agents during object detection collaboration by adversarial training. TUQCP then alleviates the impacts of adversarial attacks by providing output uncertainty estimation through learning-based module and uncertainty calibration through conformal prediction. Our framework works for early and intermediate collaboration COD models and single-agent object detection models. We evaluate TUQCP on V2X-Sim, a comprehensive collaborative perception dataset for autonomous driving, and demonstrate a 80.41% improvement in object detection accuracy compared to the baselines under the same adversarial attacks. TUQCP demonstrates the importance of uncertainty quantification to COD under adversarial attacks.

Data generation is a fundamental research problem in data management due to its diverse use cases, ranging from testing database engines to data-specific applications. However, real-world entities often involve complex interactions that cannot be effectively modeled by traditional tabular data. Therefore, graph data generation has attracted increasing attention recently. Although various graph generators have been proposed in the literature, there are three limitations: i) They cannot capture the co-evolution pattern of graph structure and node attributes. ii) Few of them consider edge direction, leading to substantial information loss. iii) Current state-of-the-art dynamic graph generators are based on the temporal random walk, making the simulation process time-consuming. To fill the research gap, we introduce VRDAG, a novel variational recurrent framework for efficient dynamic attributed graph generation. Specifically, we design a bidirectional message-passing mechanism to encode both directed structural knowledge and attribute information of a snapshot. Then, the temporal dependency in the graph sequence is captured by a recurrence state updater, generating embeddings that can preserve the evolution pattern of early graphs. Based on the hidden node embeddings, a conditional variational Bayesian method is developed to sample latent random variables at the neighboring timestep for new snapshot generation. The proposed generation paradigm avoids the time-consuming path sampling and merging process in existing random walk-based methods, significantly reducing the synthesis time. Finally, comprehensive experiments on real-world datasets are conducted to demonstrate the effectiveness and efficiency of the proposed model.

Diabetic retinopathy (DR) is one of the most common eye conditions among diabetic patients. However, vision loss occurs primarily in the late stages of DR, and the symptoms of visual impairment, ranging from mild to severe, can vary greatly, adding to the burden of diagnosis and treatment in clinical practice. Deep learning methods based on retinal images have achieved remarkable success in automatic DR grading, but most of them neglect that the presence of diabetes usually affects both eyes, and ophthalmologists usually compare both eyes concurrently for DR diagnosis, leaving correlations between left and right eyes unexploited. In this study, simulating the diagnostic process, we propose a two-stream binocular network to capture the subtle correlations between left and right eyes, in which, paired images of eyes are fed into two identical subnetworks separately during training. We design a contrastive grading loss to learn binocular correlation for five-class DR detection, which maximizes inter-class dissimilarity while minimizing the intra-class difference. Experimental results on the EyePACS dataset show the superiority of the proposed binocular model, outperforming monocular methods by a large margin.

Abstract--Sum-product networks (SPNs) represent an emerging class of neural networks with clear probabilistic semantics and superior inference speed over graphical models. This work reveals a strikingly intimate connection between SPNs and tensor networks, thus leading to a highly efficient representation that we call tensor SPNs (tSPNs). For the first time, through mapping an SPN onto a tSPN and employing novel optimization techniques, we demonstrate remarkable parameter compression with negligible loss in accuracy. INCE the inception of sum-product networks (SPNs) [1], a multitude of works have emerged with respect to their structure and weight learning, e.g., [2], [3], [4], as well as their application in image completion, speech modeling, semantic mapping and robotics, e.g., [5], just to name a few. An SPN exhibits a clear semantics of mixtures (sum nodes) and features (product nodes). Compared to other probabilistic graphical models like Bayesian and Markov networks with #P or NPhard computation, an SPN enjoys a tractable exact inference cost, and its learning is relatively simple and fast. On the other hand, there has been an exploding number of works on tensors (a multilinear operator rooted in physics) [6] including their connection and utilization in various engineering fields such as signal processing [7], and lately also in neural networks and machine learning [8], [9], [10].

There has been growing interest in extending traditional vector-based machine learning techniques to their tensor forms. An example is the support tensor machine (STM) that utilizes a rank-one tensor to capture the data structure, thereby alleviating the overfitting and curse of dimensionality problems in the conventional support vector machine (SVM). However, the expressive power of a rank-one tensor is restrictive for many real-world data. To overcome this limitation, we introduce a support tensor train machine (STTM) by replacing the rank-one tensor in an STM with a tensor train. Experiments validate and confirm the superiority of an STTM over the SVM and STM.