We address the task of controlled generation of small molecules, which entails finding novel molecules with desired properties under certain constraints (e.g., similarity to a reference molecule). Here we introduce MolMIM, a probabilistic auto-encoder for small molecule drug discovery that learns an informative and clustered latent space. MolMIM is trained with Mutual Information Machine (MIM) learning, and provides a fixed length representation of variable length SMILES strings. Since encoder-decoder models can learn representations with ``holes'' of invalid samples, here we propose a novel extension to the training procedure which promotes a dense latent space, and allows the model to sample valid molecules from random perturbations of latent codes. We provide a thorough comparison of MolMIM to several variable-size and fixed-size encoder-decoder models, demonstrating MolMIM's superior generation as measured in terms of validity, uniqueness, and novelty. We then utilize CMA-ES, a naive black-box and gradient free search algorithm, over MolMIM's latent space for the task of property guided molecule optimization. We achieve state-of-the-art results in several constrained single property optimization tasks as well as in the challenging task of multi-objective optimization, improving over previous success rate SOTA by more than 5\% . We attribute the strong results to MolMIM's latent representation which clusters similar molecules in the latent space, whereas CMA-ES is often used as a baseline optimization method. We also demonstrate MolMIM to be favourable in a compute limited regime, making it an attractive model for such cases.
Multi-task learning of dense prediction tasks, by sharing both the encoder and decoder, as opposed to sharing only the encoder, provides an attractive front to increase both accuracy and computational efficiency. When the tasks are similar, sharing the decoder serves as an additional inductive bias providing more room for tasks to share complementary information among themselves. However, increased sharing exposes more parameters to task interference which likely hinders both generalization and robustness. Effective ways to curb this interference while exploiting the inductive bias of sharing the decoder remains an open challenge. To address this challenge, we propose Progressive Decoder Fusion (PDF) to progressively combine task decoders based on inter-task representation similarity. We show that this procedure leads to a multi-task network with better generalization to in-distribution and out-of-distribution data and improved robustness to adversarial attacks. Additionally, we observe that the predictions of different tasks of this multi-task network are more consistent with each other.
Multi-modal aspect-based sentiment classification (MABSC) is task of classifying the sentiment of a target entity mentioned in a sentence and an image. However, previous methods failed to account for the fine-grained semantic association between the image and the text, which resulted in limited identification of fine-grained image aspects and opinions. To address these limitations, in this paper we propose a new approach called SeqCSG, which enhances the encoder-decoder sentiment classification framework using sequential cross-modal semantic graphs. SeqCSG utilizes image captions and scene graphs to extract both global and local fine-grained image information and considers them as elements of the cross-modal semantic graph along with tokens from tweets. The sequential cross-modal semantic graph is represented as a sequence with a multi-modal adjacency matrix indicating relationships between elements. Experimental results show that the approach outperforms existing methods and achieves state-of-the-art performance on two standard datasets. Further analysis has demonstrated that the model can implicitly learn the correlation between fine-grained information of the image and the text with the given target. Our code is available at https://github.com/zjukg/SeqCSG.
Missing data scenarios are very common in ML applications in general and time-series/sequence applications are no exceptions. This paper pertains to a novel Recurrent Neural Network (RNN) based solution for sequence prediction under missing data. Our method is distinct from all existing approaches. It tries to encode the missingness patterns in the data directly without trying to impute data either before or during model building. Our encoding is lossless and achieves compression. It can be employed for both sequence classification and forecasting. We focus on forecasting here in a general context of multi-step prediction in presence of possible exogenous inputs. In particular, we propose novel variants of Encoder-Decoder (Seq2Seq) RNNs for this. The encoder here adopts the above mentioned pattern encoding, while at the decoder which has a different structure, multiple variants are feasible. We demonstrate the utility of our proposed architecture via multiple experiments on both single and multiple sequence (real) data-sets. We consider both scenarios where (i)data is naturally missing and (ii)data is synthetically masked.
Facial expression recognition (FER) is a challenging problem because the expression component is always entangled with other irrelevant factors, such as identity and head pose. In this work, we propose an identity and pose disentangled facial expression recognition (IPD-FER) model to learn more discriminative feature representation. We regard the holistic facial representation as the combination of identity, pose and expression. These three components are encoded with different encoders. For identity encoder, a well pre-trained face recognition model is utilized and fixed during training, which alleviates the restriction on specific expression training data in previous works and makes the disentanglement practicable on in-the-wild datasets. At the same time, the pose and expression encoder are optimized with corresponding labels. Combining identity and pose feature, a neutral face of input individual should be generated by the decoder. When expression feature is added, the input image should be reconstructed. By comparing the difference between synthesized neutral and expressional images of the same individual, the expression component is further disentangled from identity and pose. Experimental results verify the effectiveness of our method on both lab-controlled and in-the-wild databases and we achieve state-of-the-art recognition performance.
Conditional graphic layout generation, which generates realistic layouts according to user constraints, is a challenging task that has not been well-studied yet. First, there is limited discussion about how to handle diverse user constraints flexibly and uniformly. Second, to make the layouts conform to user constraints, existing work often sacrifices generation quality significantly. In this work, we propose LayoutFormer++ to tackle the above problems. First, to flexibly handle diverse constraints, we propose a constraint serialization scheme, which represents different user constraints as sequences of tokens with a predefined format. Then, we formulate conditional layout generation as a sequence-to-sequence transformation, and leverage encoder-decoder framework with Transformer as the basic architecture. Furthermore, to make the layout better meet user requirements without harming quality, we propose a decoding space restriction strategy. Specifically, we prune the predicted distribution by ignoring the options that definitely violate user constraints and likely result in low-quality layouts, and make the model samples from the restricted distribution. Experiments demonstrate that LayoutFormer++ outperforms existing approaches on all the tasks in terms of both better generation quality and less constraint violation.
Prognostics and Health Management (PHM) is an emerging field which has received much attention from the manufacturing industry because of the benefits and efficiencies it brings to the table. And Remaining Useful Life (RUL) prediction is at the heart of any PHM system. Most recent data-driven research demand substantial volumes of labelled training data before a performant model can be trained under the supervised learning paradigm. This is where Transfer Learning (TL) and Domain Adaptation (DA) methods step in and make it possible for us to generalize a supervised model to other domains with different data distributions with no labelled data. In this paper, we propose \textit{LAMA-Net}, an encoder-decoder based model (Transformer) with an induced bottleneck, Latent Alignment using Maximum Mean Discrepancy (MMD) and manifold learning is proposed to tackle the problem of Unsupervised Homogeneous Domain Adaptation for RUL prediction. \textit{LAMA-Net} is validated using the C-MAPSS Turbofan Engine dataset by NASA and compared against other state-of-the-art techniques for DA. The results suggest that the proposed method offers a promising approach to perform domain adaptation in RUL prediction. Code will be made available once the paper comes out of review.
Graph anomaly detection on attributed networks has become a prevalent research topic due to its broad applications in many influential domains. In real-world scenarios, nodes and edges in attributed networks usually display distinct heterogeneity, i.e. attributes of different types of nodes show great variety, different types of relations represent diverse meanings. Anomalies usually perform differently from the majority in various perspectives of heterogeneity in these networks. However, existing graph anomaly detection approaches do not leverage heterogeneity in attributed networks, which is highly related to anomaly detection. In light of this problem, we propose AHEAD: a heterogeneity-aware unsupervised graph anomaly detection approach based on the encoder-decoder framework. Specifically, for the encoder, we design three levels of attention, i.e. attribute level, node type level, and edge level attentions to capture the heterogeneity of network structure, node properties and information of a single node, respectively. In the decoder, we exploit structure, attribute, and node type reconstruction terms to obtain an anomaly score for each node. Extensive experiments show the superiority of AHEAD on several real-world heterogeneous information networks compared with the state-of-arts in the unsupervised setting. Further experiments verify the effectiveness and robustness of our triple attention, model backbone, and decoder in general.
GitHub commits, which record the code changes with natural language messages for description, play a critical role for software developers to comprehend the software evolution. To promote the development of the open-source software community, we collect a commit benchmark including over 7.99 million commits across 7 programming languages. Based on this benchmark, we present CommitBART, a large pre-trained encoder-decoder Transformer model for GitHub commits. The model is pre-trained by three categories (i.e., denoising objectives, cross-modal generation and contrastive learning) for six pre-training tasks to learn commit fragment representations. Furthermore, we unify a ``commit intelligence'' framework with one understanding task and three generation tasks for commits. The comprehensive experiments on these tasks demonstrate that CommitBARTsignificantly outperforms previous pre-trained works for code. Further analysis also reveals each pre-training task enhances the model performance.
Dense passage retrieval aims to retrieve the relevant passages of a query from a large corpus based on dense representations (i.e., vectors) of the query and the passages. Recent studies have explored improving pre-trained language models to boost dense retrieval performance. This paper proposes CoT-MAE (ConTextual Masked Auto-Encoder), a simple yet effective generative pre-training method for dense passage retrieval. CoT-MAE employs an asymmetric encoder-decoder architecture that learns to compress the sentence semantics into a dense vector through self-supervised and context-supervised masked auto-encoding. Precisely, self-supervised masked auto-encoding learns to model the semantics of the tokens inside a text span, and context-supervised masked auto-encoding learns to model the semantical correlation between the text spans. We conduct experiments on large-scale passage retrieval benchmarks and show considerable improvements over strong baselines, demonstrating the high efficiency of CoT-MAE. Our code is available at https://github.com/caskcsg/ir/tree/main/cotmae.
Multilingual Neural Machine Translation has been showing great success using transformer models. Deploying these models is challenging because they usually require large vocabulary (vocab) sizes for various languages. This limits the speed of predicting the output tokens in the last vocab projection layer. To alleviate these challenges, this paper proposes a fast vocabulary projection method via clustering which can be used for multilingual transformers on GPUs. First, we offline split the vocab search space into disjoint clusters given the hidden context vector of the decoder output, which results in much smaller vocab columns for vocab projection. Second, at inference time, the proposed method predicts the clusters and candidate active tokens for hidden context vectors at the vocab projection. This paper also includes analysis of different ways of building these clusters in multilingual settings. Our results show end-to-end speed gains in float16 GPU inference up to 25% while maintaining the BLEU score and slightly increasing memory cost. The proposed method speeds up the vocab projection step itself by up to 2.6x. We also conduct an extensive human evaluation to verify the proposed method preserves the quality of the translations from the original model.
Deep learning models, e.g. supervised Encoder-Decoder style networks, exhibit promising performance in medical image segmentation, but come with a high labelling cost. We propose TriSegNet, a semi-supervised semantic segmentation framework. It uses triple-view feature learning on a limited amount of labelled data and a large amount of unlabeled data. The triple-view architecture consists of three pixel-level classifiers and a low-level shared-weight learning module. The model is first initialized with labelled data. Label processing, including data perturbation, confidence label voting and unconfident label detection for annotation, enables the model to train on labelled and unlabeled data simultaneously. The confidence of each model gets improved through the other two views of the feature learning. This process is repeated until each model reaches the same confidence level as its counterparts. This strategy enables triple-view learning of generic medical image datasets. Bespoke overlap-based and boundary-based loss functions are tailored to the different stages of the training. The segmentation results are evaluated on four publicly available benchmark datasets including Ultrasound, CT, MRI, and Histology images. Repeated experiments demonstrate the effectiveness of the proposed network compared against other semi-supervised algorithms, across a large set of evaluation measures.
Self-attention based transformer models have been dominating many computer vision tasks in the past few years. Their superb model qualities heavily depend on the excessively large labeled image datasets. In order to reduce the reliance on large labeled datasets, reconstruction based masked autoencoders are gaining popularity, which learn high quality transferable representations from unlabeled images. For the same purpose, recent weakly supervised image pretraining methods explore language supervision from text captions accompanying the images. In this work, we propose masked image pretraining on language assisted representation, dubbed as MILAN. Instead of predicting raw pixels or low level features, our pretraining objective is to reconstruct the image features with substantial semantic signals that are obtained using caption supervision. Moreover, to accommodate our reconstruction target, we propose a more effective prompting decoder architecture and a semantic aware mask sampling mechanism, which further advance the transfer performance of the pretrained model. Experimental results demonstrate that MILAN delivers higher accuracy than the previous works. When the masked autoencoder is pretrained and finetuned on ImageNet-1K dataset with an input resolution of 224x224, MILAN achieves a top-1 accuracy of 85.4% on ViT-Base, surpassing previous state-of-the-arts by 1%. In the downstream semantic segmentation task, MILAN achieves 52.7 mIoU using ViT-Base on ADE20K dataset, outperforming previous masked pretraining results by 4 points.
This paper proposes a simple yet effective method to improve direct (X-to-Y) translation for both cases: zero-shot and when direct data is available. We modify the input tokens at both the encoder and decoder to include signals for the source and target languages. We show a performance gain when training from scratch, or finetuning a pretrained model with the proposed setup. In the experiments, our method shows nearly 10.0 BLEU points gain on in-house datasets depending on the checkpoint selection criteria. In a WMT evaluation campaign, From-English performance improves by 4.17 and 2.87 BLEU points, in the zero-shot setting, and when direct data is available for training, respectively. While X-to-Y improves by 1.29 BLEU over the zero-shot baseline, and 0.44 over the many-to-many baseline. In the low-resource setting, we see a 1.5~1.7 point improvement when finetuning on X-to-Y domain data.
Single-step retrosynthesis is the cornerstone of retrosynthesis planning, which is a crucial task for computer-aided drug discovery. The goal of single-step retrosynthesis is to identify the possible reactants that lead to the synthesis of the target product in one reaction. By representing organic molecules as canonical strings, existing sequence-based retrosynthetic methods treat the product-to-reactant retrosynthesis as a sequence-to-sequence translation problem. However, most of them struggle to identify diverse chemical reactions for a desired product due to the deterministic inference, which contradicts the fact that many compounds can be synthesized through various reaction types with different sets of reactants. In this work, we aim to increase reaction diversity and generate various reactants using discrete latent variables. We propose a novel sequence-based approach, namely RetroDVCAE, which incorporates conditional variational autoencoders into single-step retrosynthesis and associates discrete latent variables with the generation process. Specifically, RetroDVCAE uses the Gumbel-Softmax distribution to approximate the categorical distribution over potential reactions and generates multiple sets of reactants with the variational decoder. Experiments demonstrate that RetroDVCAE outperforms state-of-the-art baselines on both benchmark dataset and homemade dataset. Both quantitative and qualitative results show that RetroDVCAE can model the multi-modal distribution over reaction types and produce diverse reactant candidates.
Hypernasality is an abnormal resonance in human speech production, especially in patients with craniofacial anomalies such as cleft palate. In clinical application, hypernasality estimation is crucial in cleft palate diagnosis, as its results determine the subsequent surgery and additional speech therapy. Therefore, designing an automatic hypernasality assessment method will facilitate speech-language pathologists to make precise diagnoses. Existing methods for hypernasality estimation only conduct acoustic analysis based on low-resource cleft palate dataset, by using statistical or neural network-based features. In this paper, we propose a novel approach that uses automatic speech recognition model to improve hypernasality estimation. Specifically, we first pre-train an encoder-decoder framework in an automatic speech recognition (ASR) objective by using speech-to-text dataset, and then fine-tune ASR encoder on the cleft palate dataset for hypernasality estimation. Benefiting from such design, our model for hypernasality estimation can enjoy the advantages of ASR model: 1) compared with low-resource cleft palate dataset, the ASR task usually includes large-scale speech data in the general domain, which enables better model generalization; 2) the text annotations in ASR dataset guide model to extract better acoustic features. Experimental results on two cleft palate datasets demonstrate that our method achieves superior performance compared with previous approaches.
Generating dense point clouds from sparse raw data benefits downstream 3D understanding tasks, but existing models are limited to a fixed upsampling ratio or to a short range of integer values. In this paper, we present APU-SMOG, a Transformer-based model for Arbitrary Point cloud Upsampling (APU). The sparse input is firstly mapped to a Spherical Mixture of Gaussians (SMOG) distribution, from which an arbitrary number of points can be sampled. Then, these samples are fed as queries to the Transformer decoder, which maps them back to the target surface. Extensive qualitative and quantitative evaluations show that APU-SMOG outperforms state-of-the-art fixed-ratio methods, while effectively enabling upsampling with any scaling factor, including non-integer values, with a single trained model. The code is available at https://github.com/apusmog/apusmog/
Renal cancer is one of the most prevalent cancers worldwide. Clinical signs of kidney cancer include hematuria and low back discomfort, which are quite distressing to the patient. Some surgery-based renal cancer treatments like laparoscopic partial nephrectomy relys on the 3D kidney parsing on computed tomography angiography (CTA) images. Many automatic segmentation techniques have been put forward to make multi-structure segmentation of the kidneys more accurate. The 3D visual model of kidney anatomy will help clinicians plan operations accurately before surgery. However, due to the diversity of the internal structure of the kidney and the low grey level of the edge. It is still challenging to separate the different parts of the kidney in a clear and accurate way. In this paper, we propose a channel extending and axial attention catching Network(CANet) for multi-structure kidney segmentation. Our solution is founded based on the thriving nn-UNet architecture. Firstly, by extending the channel size, we propose a larger network, which can provide a broader perspective, facilitating the extraction of complex structural information. Secondly, we include an axial attention catching(AAC) module in the decoder, which can obtain detailed information for refining the edges. We evaluate our CANet on the KiPA2022 dataset, achieving the dice scores of 95.8%, 89.1%, 87.5% and 84.9% for kidney, tumor, artery and vein, respectively, which helps us get fourth place in the challenge.
Error correction techniques remain effective to refine outputs from automatic speech recognition (ASR) models. Existing end-to-end error correction methods based on an encoder-decoder architecture process all tokens in the decoding phase, creating undesirable latency. In this paper, we propose an ASR error correction method utilizing the predictions of correction operations. More specifically, we construct a predictor between the encoder and the decoder to learn if a token should be kept ("K"), deleted ("D"), or changed ("C") to restrict decoding to only part of the input sequence embeddings (the "C" tokens) for fast inference. Experiments on three public datasets demonstrate the effectiveness of the proposed approach in reducing the latency of the decoding process in ASR correction. It enhances the inference speed by at least three times (3.4 and 5.7 times) while maintaining the same level of accuracy (with WER reductions of 0.53% and 1.69% respectively) for our two proposed models compared to a solid encoder-decoder baseline. In the meantime, we produce and release a benchmark dataset contributing to the ASR error correction community to foster research along this line.
Solving partial differential equations (PDEs) is an important research means in the fields of physics, biology, and chemistry. As an approximate alternative to numerical methods, PINN has received extensive attention and played an important role in many fields. However, PINN uses a fully connected network as its model, which has limited fitting ability and limited extrapolation ability in both time and space. In this paper, we propose PhyGNNet for solving partial differential equations on the basics of a graph neural network which consists of encoder, processer, and decoder blocks. In particular, we divide the computing area into regular grids, define partial differential operators on the grids, then construct pde loss for the network to optimize to build PhyGNNet model. What's more, we conduct comparative experiments on Burgers equation and heat equation to validate our approach, the results show that our method has better fit ability and extrapolation ability both in time and spatial areas compared with PINN.