Insects, such as fruit flies and honey bees, can solve simple associative learning tasks and learn abstract concepts such as "sameness" and "difference", which is viewed as a higher-order cognitive function and typically thought to depend on top-down neocortical processing. Empirical research with fruit flies strongly supports that a randomized representational architecture is used in olfactory processing in insect brains. Based on these results, we propose a Randomly Weighted Feature Network (RWFN) that incorporates randomly drawn, untrained weights in an encoder that uses an adapted linear model as a decoder. The randomized projections between input neurons and higher-order processing centers in the input brain is mimicked in RWFN by a single-hidden-layer neural network that specially structures latent representations in the hidden layer using random Fourier features that better represent complex relationships between inputs using kernel approximation. Because of this special representation, RWFNs can effectively learn the degree of relationship among inputs by training only a linear decoder model. We compare the performance of RWFNs to LTNs for Semantic Image Interpretation (SII) tasks that have been used as a representative example of how LTNs utilize reasoning over first-order logic to surpass the performance of solely data-driven methods. We demonstrate that compared to LTNs, RWFNs can achieve better or similar performance for both object classification and detection of the part-of relations between objects in SII tasks while using much far fewer learnable parameters (1:62 ratio) and a faster learning process (1:2 ratio of running speed). Furthermore, we show that because the randomized weights do not depend on the data, several decoders can share a single randomized encoder, giving RWFNs a unique economy of spatial scale for simultaneous classification tasks.
We investigate joint source channel coding (JSCC) for wireless image transmission over multipath fading channels. Inspired by recent works on deep learning based JSCC and model-based learning methods, we combine an autoencoder with orthogonal frequency division multiplexing (OFDM) to cope with multipath fading. The proposed encoder and decoder use convolutional neural networks (CNNs) and directly map the source images to complex-valued baseband samples for OFDM transmission. The multipath channel and OFDM are represented by non-trainable (deterministic) but differentiable layers so that the system can be trained end-to-end. Furthermore, our JSCC decoder further incorporates explicit channel estimation, equalization, and additional subnets to enhance the performance. The proposed method exhibits 2.5 -- 4 dB SNR gain for the equivalent image quality compared to conventional schemes that employ state-of-the-art but separate source and channel coding such as BPG and LDPC. The performance further improves when the system incorporates the channel state information (CSI) feedback. The proposed scheme is robust against OFDM signal clipping and parameter mismatch for the channel model used in training and evaluation.
Previous abstractive methods apply sequence-to-sequence structures to generate summary without a module to assist the system to detect vital mentions and relationships within a document. To address this problem, we utilize semantic graph to boost the generation performance. Firstly, we extract important entities from each document and then establish a graph inspired by the idea of distant supervision \citep{mintz-etal-2009-distant}. Then, we combine a Bi-LSTM with a graph encoder to obtain the representation of each graph node. A novel neural decoder is presented to leverage the information of such entity graphs. Automatic and human evaluations show the effectiveness of our technique.
With the rapid increase of multimedia data, a large body of literature has emerged to work on multimodal summarization, the majority of which target at refining salient information from textual and visual modalities to output a pictorial summary with the most relevant images. Existing methods mostly focus on either extractive or abstractive summarization and rely on qualified image captions to build image references. We are the first to propose a Unified framework for Multimodal Summarization grounding on BART, UniMS, that integrates extractive and abstractive objectives, as well as selecting the image output. Specially, we adopt knowledge distillation from a vision-language pretrained model to improve image selection, which avoids any requirement on the existence and quality of image captions. Besides, we introduce a visual guided decoder to better integrate textual and visual modalities in guiding abstractive text generation. Results show that our best model achieves a new state-of-the-art result on a large-scale benchmark dataset. The newly involved extractive objective as well as the knowledge distillation technique are proven to bring a noticeable improvement to the multimodal summarization task.
Open-domain dialogue generation in natural language processing (NLP) is by default a pure-language task, which aims to satisfy human need for daily communication on open-ended topics by producing related and informative responses. In this paper, we point out that hidden images, named as visual impressions (VIs), can be explored from the text-only data to enhance dialogue understanding and help generate better responses. Besides, the semantic dependency between an dialogue post and its response is complicated, e.g., few word alignments and some topic transitions. Therefore, the visual impressions of them are not shared, and it is more reasonable to integrate the response visual impressions (RVIs) into the decoder, rather than the post visual impressions (PVIs). However, both the response and its RVIs are not given directly in the test process. To handle the above issues, we propose a framework to explicitly construct VIs based on pure-language dialogue datasets and utilize them for better dialogue understanding and generation. Specifically, we obtain a group of images (PVIs) for each post based on a pre-trained word-image mapping model. These PVIs are used in a co-attention encoder to get a post representation with both visual and textual information. Since the RVIs are not provided directly during testing, we design a cascade decoder that consists of two sub-decoders. The first sub-decoder predicts the content words in response, and applies the word-image mapping model to get those RVIs. Then, the second sub-decoder generates the response based on the post and RVIs. Experimental results on two open-domain dialogue datasets show that our proposed approach achieves superior performance over competitive baselines.
In this paper, we take the advantage of previous pre-trained models (PTMs) and propose a novel Chinese Pre-trained Unbalanced Transformer (CPT). Different from previous Chinese PTMs, CPT is designed to utilize the shared knowledge between natural language understanding (NLU) and natural language generation (NLG) to boost the performance. CPT consists of three parts: a shared encoder, an understanding decoder, and a generation decoder. Two specific decoders with a shared encoder are pre-trained with masked language modeling (MLM) and denoising auto-encoding (DAE) tasks, respectively. With the partially shared architecture and multi-task pre-training, CPT can (1) learn specific knowledge of both NLU or NLG tasks with two decoders and (2) be fine-tuned flexibly that fully exploits the potential of the model. Moreover, the unbalanced Transformer saves the computational and storage cost, which makes CPT competitive and greatly accelerates the inference of text generation. Experimental results on a wide range of Chinese NLU and NLG tasks show the effectiveness of CPT.
We present an algorithm for re-rendering a person from a single image under arbitrary poses. Existing methods often have difficulties in hallucinating occluded contents photo-realistically while preserving the identity and fine details in the source image. We first learn to inpaint the correspondence field between the body surface texture and the source image with a human body symmetry prior. The inpainted correspondence field allows us to transfer/warp local features extracted from the source to the target view even under large pose changes. Directly mapping the warped local features to an RGB image using a simple CNN decoder often leads to visible artifacts. Thus, we extend the StyleGAN generator so that it takes pose as input (for controlling poses) and introduces a spatially varying modulation for the latent space using the warped local features (for controlling appearances). We show that our method compares favorably against the state-of-the-art algorithms in both quantitative evaluation and visual comparison.
Video grounding aims to localize the temporal segment corresponding to a sentence query from an untrimmed video. Almost all existing video grounding methods fall into two frameworks: 1) Top-down model: It predefines a set of segment candidates and then conducts segment classification and regression. 2) Bottom-up model: It directly predicts frame-wise probabilities of the referential segment boundaries. However, all these methods are not end-to-end, i.e., they always rely on some time-consuming post-processing steps to refine predictions. To this end, we reformulate video grounding as a set prediction task and propose a novel end-to-end multi-modal Transformer model, dubbed as GTR. Specifically, GTR has two encoders for video and language encoding, and a cross-modal decoder for grounding prediction. To facilitate the end-to-end training, we use a Cubic Embedding layer to transform the raw videos into a set of visual tokens. To better fuse these two modalities in the decoder, we design a new Multi-head Cross-Modal Attention. The whole GTR is optimized via a Many-to-One matching loss. Furthermore, we conduct comprehensive studies to investigate different model design choices. Extensive results on three benchmarks have validated the superiority of GTR. All three typical GTR variants achieve record-breaking performance on all datasets and metrics, with several times faster inference speed.
Pixel-level road crack detection has always been a challenging task in intelligent transportation systems. Due to the external environments, such as weather, light, and other factors, pavement cracks often present low contrast, poor continuity, and different sizes in length and width. However, most of the existing studies pay less attention to crack data under different situations. Meanwhile, recent algorithms based on deep convolutional neural networks (DCNNs) have promoted the development of cutting-edge models for crack detection. Nevertheless, they usually focus on complex models for good performance, but ignore detection efficiency in practical applications. In this article, to address the first issue, we collected two new databases (i.e. Rain365 and Sun520) captured in rainy and sunny days respectively, which enrich the data of the open source community. For the second issue, we reconsider how to improve detection efficiency with excellent performance, and then propose our lightweight encoder-decoder architecture termed CarNet. Specifically, we introduce a novel olive-shaped structure for the encoder network, a light-weight multi-scale block and a new up-sampling method in the decoder network. Numerous experiments show that our model can better balance detection performance and efficiency compared with previous models. Especially, on the Sun520 dataset, our CarNet significantly advances the state-of-the-art performance with ODS F-score from 0.488 to 0.514. Meanwhile, it does so with an improved detection speed (104 frame per second) which is orders of magnitude faster than some recent DCNNs-based algorithms specially designed for crack detection.
The domain gap caused mainly by variable medical image quality renders a major obstacle on the path between training a segmentation model in the lab and applying the trained model to unseen clinical data. To address this issue, domain generalization methods have been proposed, which however usually use static convolutions and are less flexible. In this paper, we propose a multi-source domain generalization model based on the domain and content adaptive convolution (DCAC) for the segmentation of medical images across different modalities. Specifically, we design the domain adaptive convolution (DAC) module and content adaptive convolution (CAC) module and incorporate both into an encoder-decoder backbone. In the DAC module, a dynamic convolutional head is conditioned on the predicted domain code of the input to make our model adapt to the unseen target domain. In the CAC module, a dynamic convolutional head is conditioned on the global image features to make our model adapt to the test image. We evaluated the DCAC model against the baseline and four state-of-the-art domain generalization methods on the prostate segmentation, COVID-19 lesion segmentation, and optic cup/optic disc segmentation tasks. Our results not only indicate that the proposed DCAC model outperforms all competing methods on each segmentation task but also demonstrate the effectiveness of the DAC and CAC modules. Code is available at \url{https://git.io/DCAC}.
We introduce Multiresolution Deep Implicit Functions (MDIF), a hierarchical representation that can recover fine geometry detail, while being able to perform global operations such as shape completion. Our model represents a complex 3D shape with a hierarchy of latent grids, which can be decoded into different levels of detail and also achieve better accuracy. For shape completion, we propose latent grid dropout to simulate partial data in the latent space and therefore defer the completing functionality to the decoder side. This along with our multires design significantly improves the shape completion quality under decoder-only latent optimization. To the best of our knowledge, MDIF is the first deep implicit function model that can at the same time (1) represent different levels of detail and allow progressive decoding; (2) support both encoder-decoder inference and decoder-only latent optimization, and fulfill multiple applications; (3) perform detailed decoder-only shape completion. Experiments demonstrate its superior performance against prior art in various 3D reconstruction tasks.
Brain-inspired computation and information processing alongside compatibility with neuromorphic hardware have made spiking neural networks (SNN) a promising method for solving learning tasks in machine learning (ML). Spiking neurons are only one of the requirements for building a bio-plausible learning model. Network architecture and learning rules are other important factors to consider when developing such artificial agents. In this work, inspired by the human visual pathway and the role of dopamine in learning, we propose a reward-modulated locally connected spiking neural network, BioLCNet, for visual learning tasks. To extract visual features from Poisson-distributed spike trains, we used local filters that are more analogous to the biological visual system compared to convolutional filters with weight sharing. In the decoding layer, we applied a spike population-based voting scheme to determine the decision of the network. We employed Spike-timing-dependent plasticity (STDP) for learning the visual features, and its reward-modulated variant (R-STDP) for training the decoder based on the reward or punishment feedback signal. For evaluation, we first assessed the robustness of our rewarding mechanism to varying target responses in a classical conditioning experiment. Afterwards, we evaluated the performance of our network on image classification tasks of MNIST and XOR MNIST datasets.
Image captioning using Encoder-Decoder based approach where CNN is used as the Encoder and sequence generator like RNN as Decoder has proven to be very effective. However, this method has a drawback that is sequence needs to be processed in order. To overcome this drawback some researcher has utilized the Transformer model to generate captions from images using English datasets. However, none of them generated captions in Bengali using the transformer model. As a result, we utilized three different Bengali datasets to generate Bengali captions from images using the Transformer model. Additionally, we compared the performance of the transformer-based model with a visual attention-based Encoder-Decoder approach. Finally, we compared the result of the transformer-based model with other models that employed different Bengali image captioning datasets.
Eliciting knowledge from pre-trained language models via prompt-based learning has shown great potential in many natural language processing tasks. Whereas, the applications for more complex tasks such as event extraction are less studied since the design of prompt is not straightforward for the structured event containing various triggers and arguments. % Meanwhile, current conditional generation methods employ large encoder-decoder models, which are costly to train and serve. In this paper, we present a novel prompt-based approach, which elicits both the independent and joint knowledge about different events for event argument extraction. The experimental results on the benchmark ACE2005 dataset show the great advantages of our proposed approach. In particular, our approach is superior to the recent advanced methods in both fully-supervised and low-resource scenarios.
This work proposes an extensive analysis of the Transformer architecture in the Neural Machine Translation (NMT) setting. Focusing on the encoder-decoder attention mechanism, we prove that attention weights systematically make alignment errors by relying mainly on uninformative tokens from the source sequence. However, we observe that NMT models assign attention to these tokens to regulate the contribution in the prediction of the two contexts, the source and the prefix of the target sequence. We provide evidence about the influence of wrong alignments on the model behavior, demonstrating that the encoder-decoder attention mechanism is well suited as an interpretability method for NMT. Finally, based on our analysis, we propose methods that largely reduce the word alignment error rate compared to standard induced alignments from attention weights.
Given a piece of speech and its transcript text, text-based speech editing aims to generate speech that can be seamlessly inserted into the given speech by editing the transcript. Existing methods adopt a two-stage approach: synthesize the input text using a generic text-to-speech (TTS) engine and then transform the voice to the desired voice using voice conversion (VC). A major problem of this framework is that VC is a challenging problem which usually needs a moderate amount of parallel training data to work satisfactorily. In this paper, we propose a one-stage context-aware framework to generate natural and coherent target speech without any training data of the target speaker. In particular, we manage to perform accurate zero-shot duration prediction for the inserted text. The predicted duration is used to regulate both text embedding and speech embedding. Then, based on the aligned cross-modality input, we directly generate the mel-spectrogram of the edited speech with a transformer-based decoder. Subjective listening tests show that despite the lack of training data for the speaker, our method has achieved satisfactory results. It outperforms a recent zero-shot TTS engine by a large margin.
Can pre-trained BERT for one language and GPT for another be glued together to translate texts? Self-supervised training using only monolingual data has led to the success of pre-trained (masked) language models in many NLP tasks. However, directly connecting BERT as an encoder and GPT as a decoder can be challenging in machine translation, for GPT-like models lack a cross-attention component that is needed in seq2seq decoders. In this paper, we propose Graformer to graft separately pre-trained (masked) language models for machine translation. With monolingual data for pre-training and parallel data for grafting training, we maximally take advantage of the usage of both types of data. Experiments on 60 directions show that our method achieves average improvements of 5.8 BLEU in x2en and 2.9 BLEU in en2x directions comparing with the multilingual Transformer of the same size.
In recent years, several high-performance conversational systems have been proposed based on the Transformer encoder-decoder model. Although previous studies analyzed the effects of the model parameters and the decoding method on subjective dialogue evaluations with overall metrics, they did not analyze how the differences of fine-tuning datasets affect on user's detailed impression. In addition, the Transformer-based approach has only been verified for English, not for such languages with large inter-language distances as Japanese. In this study, we develop large-scale Transformer-based Japanese dialogue models and Japanese chit-chat datasets to examine the effectiveness of the Transformer-based approach for building chit-chat dialogue systems. We evaluated and analyzed the impressions of human dialogues in different fine-tuning datasets, model parameters, and the use of additional information.
Probabilistic generative models are attractive for scientific modeling because their inferred parameters can be used to generate hypotheses and design experiments. This requires that the learned model provide an accurate representation of the input data and yield a latent space that effectively predicts outcomes relevant to the scientific question. Supervised Variational Autoencoders (SVAEs) have previously been used for this purpose, where a carefully designed decoder can be used as an interpretable generative model while the supervised objective ensures a predictive latent representation. Unfortunately, the supervised objective forces the encoder to learn a biased approximation to the generative posterior distribution, which renders the generative parameters unreliable when used in scientific models. This issue has remained undetected as reconstruction losses commonly used to evaluate model performance do not detect bias in the encoder. We address this previously-unreported issue by developing a second order supervision framework (SOS-VAE) that influences the decoder to induce a predictive latent representation. This ensures that the associated encoder maintains a reliable generative interpretation. We extend this technique to allow the user to trade-off some bias in the generative parameters for improved predictive performance, acting as an intermediate option between SVAEs and our new SOS-VAE. We also use this methodology to address missing data issues that often arise when combining recordings from multiple scientific experiments. We demonstrate the effectiveness of these developments using synthetic data and electrophysiological recordings with an emphasis on how our learned representations can be used to design scientific experiments.
Multilingual Neural Machine Translation (NMT) enables one model to serve all translation directions, including ones that are unseen during training, i.e. zero-shot translation. Despite being theoretically attractive, current models often produce low quality translations -- commonly failing to even produce outputs in the right target language. In this work, we observe that off-target translation is dominant even in strong multilingual systems, trained on massive multilingual corpora. To address this issue, we propose a joint approach to regularize NMT models at both representation-level and gradient-level. At the representation level, we leverage an auxiliary target language prediction task to regularize decoder outputs to retain information about the target language. At the gradient level, we leverage a small amount of direct data (in thousands of sentence pairs) to regularize model gradients. Our results demonstrate that our approach is highly effective in both reducing off-target translation occurrences and improving zero-shot translation performance by +5.59 and +10.38 BLEU on WMT and OPUS datasets respectively. Moreover, experiments show that our method also works well when the small amount of direct data is not available.