ICASSP2023 General Meeting Understanding and Generation Challenge (MUG) focuses on prompting a wide range of spoken language processing (SLP) research on meeting transcripts, as SLP applications are critical to improve users' efficiency in grasping important information in meetings. MUG includes five tracks, including topic segmentation, topic-level and session-level extractive summarization, topic title generation, keyphrase extraction, and action item detection. To facilitate MUG, we construct and release a large-scale meeting dataset, the AliMeeting4MUG Corpus.

Learning on a massive amount of speech corpus leads to the recent success of many self-supervised speech models. With knowledge distillation, these models may also benefit from the knowledge encoded by language models that are pre-trained on rich sources of texts. The distillation process, however, is challenging due to the modal disparity between textual and speech embedding spaces. This paper studies metric-based distillation to align the embedding space of text and speech with only a small amount of data without modifying the model structure. Since the semantic and granularity gap between text and speech has been omitted in literature, which impairs the distillation, we propose the Prior-informed Adaptive knowledge Distillation (PAD) that adaptively leverages text/speech units of variable granularity and prior distributions to achieve better global and local alignments between text and speech pre-trained models. We evaluate on three spoken language understanding benchmarks to show that PAD is more effective in transferring linguistic knowledge than other metric-based distillation approaches.

We launch EVA, a vision-centric foundation model to explore the limits of visual representation at scale using only publicly accessible data. EVA is a vanilla ViT pre-trained to reconstruct the masked out image-text aligned vision features conditioned on visible image patches. Via this pretext task, we can efficiently scale up EVA to one billion parameters, and sets new records on a broad range of representative vision downstream tasks, such as image recognition, video action recognition, object detection, instance segmentation and semantic segmentation without heavy supervised training. Moreover, we observe quantitative changes in scaling EVA result in qualitative changes in transfer learning performance that are not present in other models. For instance, EVA takes a great leap in the challenging large vocabulary instance segmentation task: our model achieves almost the same state-of-the-art performance on LVISv1.0 dataset with over a thousand categories and COCO dataset with only eighty categories. Beyond a pure vision encoder, EVA can also serve as a vision-centric, multi-modal pivot to connect images and text. We find initializing the vision tower of a giant CLIP from EVA can greatly stabilize the training and outperform the training from scratch counterpart with much fewer samples and less compute, providing a new direction for scaling up and accelerating the costly training of multi-modal foundation models. To facilitate future research, we release all the code and models at https://github.com/baaivision/EVA.

Detection transformers have recently shown promising object detection results and attracted increasing attention. However, how to develop effective domain adaptation techniques to improve its cross-domain performance remains unexplored and unclear. In this paper, we delve into this topic and empirically find that direct feature distribution alignment on the CNN backbone only brings limited improvements, as it does not guarantee domain-invariant sequence features in the transformer for prediction. To address this issue, we propose a novel Sequence Feature Alignment (SFA) method that is specially designed for the adaptation of detection transformers. Technically, SFA consists of a domain query-based feature alignment (DQFA) module and a token-wise feature alignment (TDA) module. In DQFA, a novel domain query is used to aggregate and align global context from the token sequence of both domains. DQFA reduces the domain discrepancy in global feature representations and object relations when deploying in the transformer encoder and decoder, respectively. Meanwhile, TDA aligns token features in the sequence from both domains, which reduces the domain gaps in local and instance-level feature representations in the transformer encoder and decoder, respectively. Besides, a novel bipartite matching consistency loss is proposed to enhance the feature discriminability for robust object detection. Experiments on three challenging benchmarks show that SFA outperforms state-of-the-art domain adaptive object detection methods. Code has been made available at: https://github.com/encounter1997/SFA.

The exemplar-free class incremental learning requires classification models to learn new class knowledge incrementally without retaining any old samples. Recently, the framework based on parallel one-class classifiers (POC), which trains a one-class classifier (OCC) independently for each category, has attracted extensive attention, since it can naturally avoid catastrophic forgetting. POC, however, suffers from weak discriminability and comparability due to its independent training strategy for different OOCs. To meet this challenge, we propose a new framework, named Discriminative and Comparable One-class classifiers for Incremental Learning (DisCOIL). DisCOIL follows the basic principle of POC, but it adopts variational auto-encoders (VAE) instead of other well-established one-class classifiers (e.g. deep SVDD), because a trained VAE can not only identify the probability of an input sample belonging to a class but also generate pseudo samples of the class to assist in learning new tasks. With this advantage, DisCOIL trains a new-class VAE in contrast with the old-class VAEs, which forces the new-class VAE to reconstruct better for new-class samples but worse for the old-class pseudo samples, thus enhancing the comparability. Furthermore, DisCOIL introduces a hinge reconstruction loss to ensure the discriminability. We evaluate our method extensively on MNIST, CIFAR10, and Tiny-ImageNet. The experimental results show that DisCOIL achieves state-of-the-art performance.

Identifying homogeneous groups of regression coefficients has received increasing attention because the resulting regression model provides better scientific interpretations and enhance predictive performance in many applications. In some occasions, features or covariates naturally act in groups to influence outcomes, so knowing group structures of the features help scientists gain new knowledge about a physical system of interest. From a modeling perspective, aggregating covariates with similar effects along with the response reduces model complexity and improves interpretability, especially in the highdimensional regime. There have been a flurry of works under this direction; see for example Bondell and Reich (2008); Shen and Huang (2010); Zhu, Shen and Pan (2013); Ke, Fan and Wu (2015); Jeon, Kwon and Choi (2017), among others. There is a vast literature in discovering homogeneous groups of observations or individuals in overly heterogeneous population. However, these existing methods cannot be applied to our problem that aims to group regression parameters.

Keyphrases are phrases in a document providing a concise summary of core content, helping readers to understand what the article is talking about in a minute. However, existing unsupervised works are not robust enough to handle various types of documents owing to the mismatch of sequence length for comparison. In this paper, we propose a novel unsupervised keyword extraction method by leveraging the BERT-based model to select and rank candidate keyphrases with a MASK strategy. In addition, we further enhance the model, denoted as Keyphrases Extraction BERT (KPEBERT), via designing a compatible self-supervised task and conducting a contrast learning. We conducted extensive experimental evaluation to demonstrate the superiority and robustness of the proposed method as well as the effectiveness of KPEBERT.

Transformer-based models have achieved great success in various NLP, vision, and speech tasks. However, the core of Transformer, the self-attention mechanism, has a quadratic time and memory complexity with respect to the sequence length, which hinders applications of Transformer-based models to long sequences. Many approaches have been proposed to mitigate this problem, such as sparse attention mechanisms, low-rank matrix approximations and scalable kernels, and token mixing alternatives to self-attention. We propose a novel Pooling Network (PoNet) for token mixing in long sequences with linear complexity. We design multi-granularity pooling and pooling fusion to capture different levels of contextual information and combine their interactions with tokens. On the Long Range Arena benchmark, PoNet significantly outperforms Transformer and achieves competitive accuracy, while being only slightly slower than the fastest model, FNet, across all sequence lengths measured on GPUs. We also conduct systematic studies on the transfer learning capability of PoNet and observe that PoNet achieves 96.0% of the accuracy of BERT on the GLUE benchmark, outperforming FNet by 4.5% relative. Comprehensive ablation analysis demonstrates effectiveness of the designed multi-granularity pooling and pooling fusion for token mixing in long sequences and efficacy of the designed pre-training tasks for PoNet to learn transferable contextualized language representations.

A table arranging data in rows and columns is a very effective data structure, which has been widely used in business and scientific research. Considering large-scale tabular data in online and offline documents, automatic table recognition has attracted increasing attention from the document analysis community. Though human can easily understand the structure of tables, it remains a challenge for machines to understand that, especially due to a variety of different table layouts and styles. Existing methods usually model a table as either the markup sequence or the adjacency matrix between different table cells, failing to address the importance of the logical location of table cells, e.g., a cell is located in the first row and the second column of the table. In this paper, we reformulate the problem of table structure recognition as the table graph reconstruction, and propose an end-to-end trainable table graph reconstruction network (TGRNet) for table structure recognition. Specifically, the proposed method has two main branches, a cell detection branch and a cell logical location branch, to jointly predict the spatial location and the logical location of different cells. Experimental results on three popular table recognition datasets and a new dataset with table graph annotations (TableGraph-350K) demonstrate the effectiveness of the proposed TGRNet for table structure recognition. Code and annotations will be made publicly available.

Stratifying patients at risk for postoperative complications may facilitate timely and accurate workups and reduce the burden of adverse events on patients and the health system. Currently, a widely-used surgical risk calculator created by the American College of Surgeons, NSQIP, uses 21 preoperative covariates to assess risk of postoperative complications, but lacks dynamic, real-time capabilities to accommodate postoperative information. We propose a new Hidden Markov Model sequence classifier for analyzing patients' postoperative temperature sequences that incorporates their time-invariant characteristics in both transition probability and initial state probability in order to develop a postoperative "real-time" complication detector. Data from elective Colectomy surgery indicate that our method has improved classification performance compared to 8 other machine learning classifiers when using the full temperature sequence associated with the patients' length of stay. Additionally, within 44 hours after surgery, the performance of the model is close to that of full-length temperature sequence.