Modern language models mostly take sub-words as input, a design that balances the trade-off between vocabulary size, number of parameters, and performance. However, sub-word tokenization still has disadvantages like not being robust to noise and difficult to generalize to new languages. Also, the current trend of scaling up models reveals that larger models require larger embeddings but that makes parallelization hard. Previous work on image classification proves splitting raw input into a sequence of chucks is a strong, model-agnostic inductive bias. Based on this observation, we rethink the existing character-aware method that takes character-level inputs but makes word-level sequence modeling and prediction. We overhaul this method by introducing a cross-attention network that builds word-level representation directly from bytes, and a sub-word level prediction based on word-level hidden states to avoid the time and space requirement of word-level prediction. With these two improvements combined, we have a token free model with slim input embeddings for downstream tasks. We name our method Byte2Word and perform evaluations on language modeling and text classification. Experiments show that Byte2Word is on par with the strong sub-word baseline BERT but only takes up 10\% of embedding size. We further test our method on synthetic noise and cross-lingual transfer and find it competitive to baseline methods on both settings.
The proposed shopping assistant model SANIP is going to help blind persons to detect hand held objects and also to get a video feedback of the information retrieved from the detected and recognized objects. The proposed model consists of three python models i.e. Custom Object Detection, Text Detection and Barcode detection. For object detection of the hand held object, we have created our own custom dataset that comprises daily goods such as Parle-G, Tide, and Lays. Other than that we have also collected images of Cart and Exit signs as it is essential for any person to use a cart and also notice the exit sign in case of emergency. For the other 2 models proposed the text and barcode information retrieved is converted from text to speech and relayed to the Blind person. The model was used to detect objects that were trained on and was successful in detecting and recognizing the desired output with a good accuracy and precision.
Scene text recognition (STR) is an important bridge between images and text, attracting abundant research attention. While convolutional neural networks (CNNS) have achieved remarkable progress in this task, most of the existing works need an extra module (context modeling module) to help CNN to capture global dependencies to solve the inductive bias and strengthen the relationship between text features. Recently, the transformer has been proposed as a promising network for global context modeling by self-attention mechanism, but one of the main shortcomings, when applied to recognition, is the efficiency. We propose a 1-D split to address the challenges of complexity and replace the CNN with the transformer encoder to reduce the need for a context modeling module. Furthermore, recent methods use a frozen initial embedding to guide the decoder to decode the features to text, leading to a loss of accuracy. We propose to use a learnable initial embedding learned from the transformer encoder to make it adaptive to different input images. Above all, we introduce a novel architecture for text recognition, named TRansformer-based text recognizer with Initial embedding Guidance (TRIG), composed of three stages (transformation, feature extraction, and prediction). Extensive experiments show that our approach can achieve state-of-the-art on text recognition benchmarks.
Consider a scenario in one-shot query-guided object localization where neither an image of the object nor the object category name is available as a query. In such a scenario, a hand-drawn sketch of the object could be a choice for a query. However, hand-drawn crude sketches alone, when used as queries, might be ambiguous for object localization, e.g., a sketch of a laptop could be confused for a sofa. On the other hand, a linguistic definition of the category, e.g., a small portable computer small enough to use in your lap" along with the sketch query, gives better visual and semantic cues for object localization. In this work, we present a multimodal query-guided object localization approach under the challenging open-set setting. In particular, we use queries from two modalities, namely, hand-drawn sketch and description of the object (also known as gloss), to perform object localization. Multimodal query-guided object localization is a challenging task, especially when a large domain gap exists between the queries and the natural images, as well as due to the challenge of combining the complementary and minimal information present across the queries. For example, hand-drawn crude sketches contain abstract shape information of an object, while the text descriptions often capture partial semantic information about a given object category. To address the aforementioned challenges, we present a novel cross-modal attention scheme that guides the region proposal network to generate object proposals relevant to the input queries and a novel orthogonal projection-based proposal scoring technique that scores each proposal with respect to the queries, thereby yielding the final localization results. ...
Recently, vector-quantized image modeling has demonstrated impressive performance on generation tasks such as text-to-image generation. However, we discover that the current image quantizers do not satisfy translation equivariance in the quantized space due to aliasing, degrading performance in the downstream text-to-image generation and image-to-text generation, even in simple experimental setups. Instead of focusing on anti-aliasing, we take a direct approach to encourage translation equivariance in the quantized space. In particular, we explore a desirable property of image quantizers, called 'Translation Equivariance in the Quantized Space' and propose a simple but effective way to achieve translation equivariance by regularizing orthogonality in the codebook embedding vectors. Using this method, we improve accuracy by +22% in text-to-image generation and +26% in image-to-text generation, outperforming the VQGAN.
Generating natural and informative texts has been a long-standing problem in NLP. Much effort has been dedicated into incorporating pre-trained language models (PLMs) with various open-world knowledge, such as knowledge graphs or wiki pages. However, their ability to access and manipulate the task-specific knowledge is still limited on downstream tasks, as this type of knowledge is usually not well covered in PLMs and is hard to acquire. To address the problem, we propose augmenting TExt Generation via Task-specific and Open-world Knowledge (TegTok) in a unified framework. Our model selects knowledge entries from two types of knowledge sources through dense retrieval and then injects them into the input encoding and output decoding stages respectively on the basis of PLMs. With the help of these two types of knowledge, our model can learn what and how to generate. Experiments on two text generation tasks of dialogue generation and question generation, and on two datasets show that our method achieves better performance than various baseline models.
We explore how to crawl financial forum data such as stock bars and combine them with deep learning models for sentiment analysis. In this paper, we will use the BERT model to train against the financial corpus and predict the SZSE Component Index, and find that applying the BERT model to the financial corpus through the maximum information coefficient comparison study. The obtained sentiment features will be able to reflect the fluctuations in the stock market and help to improve the prediction accuracy effectively. Meanwhile, this paper combines deep learning with financial text, in further exploring the mechanism of investor sentiment on stock market through deep learning method, which will be beneficial for national regulators and policy departments to develop more reasonable policy guidelines for maintaining the stability of stock market.
In this paper, we introduce our work of building a Streaming Multilingual Speech Model (SM2), which can transcribe or translate multiple spoken languages into texts of the target language. The backbone of SM2 is Transformer Transducer, which has high streaming capability. Instead of human labeled speech translation (ST) data, SM2 models are trained using weakly supervised data generated by converting the transcriptions in speech recognition corpora with a machine translation service. With 351 thousand hours of anonymized speech training data from 25 languages, SM2 models achieve comparable or even better ST quality than some recent popular large-scale non-streaming speech models. More importantly, we show that SM2 has the truly zero-shot capability when expanding to new target languages, yielding high quality ST results for {source-speech, target-text} pairs that are not seen during training.
The Longest Common Subsequence (LCS) is the problem of finding a subsequence among a set of strings that has two properties of being common to all and is the longest. The LCS has applications in computational biology and text editing, among many others. Due to the NP-hardness of the general longest common subsequence, numerous heuristic algorithms and solvers have been proposed to give the best possible solution for different sets of strings. None of them has the best performance for all types of sets. In addition, there is no method to specify the type of a given set of strings. Besides that, the available hyper-heuristic is not efficient and fast enough to solve this problem in real-world applications. This paper proposes a novel hyper-heuristic to solve the longest common subsequence problem using a novel criterion to classify a set of strings based on their similarity. To do this, we offer a general stochastic framework to identify the type of a given set of strings. Following that, we introduce the set similarity dichotomizer ($S^2D$) algorithm based on the framework that divides the type of sets into two. This algorithm is introduced for the first time in this paper and opens a new way to go beyond the current LCS solvers. Then, we present a novel hyper-heuristic that exploits the $S^2D$ and one of the internal properties of the set to choose the best matching heuristic among a set of heuristics. We compare the results on benchmark datasets with the best heuristics and hyper-heuristics. The results show a higher performance of our proposed hyper-heuristic in both quality of solutions and run time factors.
The emergence of Long Short-Term Memory (LSTM) solves the problems of vanishing gradient and exploding gradient in traditional Recurrent Neural Networks (RNN). LSTM, as a new type of RNN, has been widely used in various fields, such as text prediction, Wind Speed Forecast, depression prediction by EEG signals, etc. The results show that improving the efficiency of LSTM can help to improve the efficiency in other application areas. In this paper, we proposed an advanced LSTM algorithm, the Extreme Long Short-Term Memory (E-LSTM), which adds the inverse matrix part of Extreme Learning Machine (ELM) as a new "gate" into the structure of LSTM. This "gate" preprocess a portion of the data and involves the processed data in the cell update of the LSTM to obtain more accurate data with fewer training rounds, thus reducing the overall training time. In this research, the E-LSTM model is used for the text prediction task. Experimental results showed that the E-LSTM sometimes takes longer to perform a single training round, but when tested on a small data set, the new E-LSTM requires only 2 epochs to obtain the results of the 7th epoch traditional LSTM. Therefore, the E-LSTM retains the high accuracy of the traditional LSTM, whilst also improving the training speed and the overall efficiency of the LSTM.