This paper studies the problem of generalized zero-shot learning which requires model to train on image-label pairs from some seen classes and test task classifying new images both unseen classes. In this paper, we propose a novel that provides unified framework for three different approaches: visual → semantic mapping, metric learning. Specifically, our proposed consists feature generator can generate various features given class embeddings as input, regressor maps each back its...

Pre-training models have been proved effective for a wide range of natural language processing tasks. Inspired by this, we propose novel dialogue generation pre-training framework to support various kinds conversations, including chit-chat, knowledge grounded dialogues, and conversational question answering. In this framework, adopt flexible attention mechanisms fully leverage the bi-directional context uni-directional characteristic generation. We also introduce discrete latent variables...

There has been a drastic growth of research in Generative Adversarial Nets (GANs) the past few years. Proposed 2014, GAN applied to various applications such as computer vision and natural language processing, achieves impressive performance. Among many GAN, image synthesis is most well-studied one, this area already demonstrated great potential using synthesis. In paper, we provide taxonomy methods used synthesis, review different models for text-to-image image-to-image translation, discuss...

To build a high-quality open-domain chatbot, we introduce the effective training process of PLATO-2 via curriculum learning.There are two stages involved in learning process.In first stage, coarse-grained generation model is trained to learn response under simplified framework oneto-one mapping.In second finegrained generative augmented with latent variables and an evaluation further generate diverse responses select best response, respectively.PLATO-2 was on both Chinese English data, whose...

Traffic event cognition and reasoning in videos is an important task that has a wide range of applications intelligent transportation, assisted driving, autonomous vehicles. In this paper, we create novel dataset, SUTD-TrafficQA (Traffic Question Answering), which takes the form video QA based on collected 10,080 in-the-wild annotated 62,535 pairs, for benchmarking cognitive capability causal inference understanding models complex traffic scenarios. Specifically, propose 6 challenging tasks...

Effective maize and weed detection plays an important role in farmland management, which helps to improve yield save herbicide resources. Due their convenience high resolution, Unmanned Aerial Vehicles (UAVs) are widely used detection. However, there some challenging problems detection: (i) the cost of labeling is high, image contains many plants, annotation time-consuming labor-intensive; (ii) number much larger than field, this imbalance samples leads decreased recognition accuracy; (iii)...

ABSTRACT Traditionally, researchers employ human raters for scoring responses to creative thinking tasks. Apart from the associated costs this approach entails two potential risks. First, can be subjective in their behavior (inter‐rater‐variance). Second, individual are prone inconsistent patterns (intra‐rater‐variance). In light of these issues, we present an automated Divergent Thinking (DT) Tasks. We implemented a pipeline aiming generate accurate rating predictions DT using text mining...

Light traps have been widely used for automatic monitoring of pests in the field as an alternative to time-consuming and labor-intensive manual investigations. However, scale variation, complex background dense distribution light-trap images bring challenges rapid accurate detection when utilizing vision technology. To overcome these challenges, this paper, we put forward a lightweight pest model, AgriPest-YOLO, achieving well-balanced between efficiency, accuracy model size detection....

The Track-1 of DSTC9 aims to effectively answer user requests or questions during task-oriented dialogues, which are out the scope APIs/DB. By leveraging external knowledge resources, relevant information can be retrieved and encoded into response generation for these out-of-API-coverage queries. In this work, we have explored several advanced techniques enhance utilization boost quality generation, including schema guided decision, negatives enhanced selection, grounded generation. To...

The application of air-ground collaborative network has become increasingly widespread in intelligent vehicular systems. In order to effectively utilize multiple unmanned aerial vehicles (UAVs) provide fast services and improve resource allocation for network, this paper proposes a 3D terrain-oriented path planning algorithm multi-UAVs assisted based on swarm intelligence optimization. It is aimed address UAVs' complex terrains, requiring substantial computation. However, the current...

Efficient navigation of multiple autonomous underwater vehicles (AUVs) plays an important role in monitoring and off-shore environments. It has encountered challenges when AUVs work complex Traditional swarm intelligence (SI) optimization algorithms have limitations such as insufficient path exploration ability, susceptibility to local optima, difficulty convergence. To address these issues, we propose improved multi-objective manta ray foraging (IMMRFO) method, which can improve the...

Traumatic pneumothorax is a complex condition that challenging to diagnose, particularly in hospitals, underdeveloped areas, and during mass casualty events. This study aimed evaluate the potential of machine learning (ML) for diagnosing assessing traumatic pneumothorax. We extracted 33 vital signs blood gas parameters from MIMIC-IV database, selecting 12 clinically significant features as inputs four ML algorithms: extreme gradient boosting (XGBoost), artificial neural network (ANN),...

Protein pre-training has emerged as a transformative approach for solving diverse biological tasks. While many contemporary methods focus on sequence-based language models, recent findings highlight that protein sequences alone are insufficient to capture the extensive information inherent in structures. Recognizing crucial role of structure defining function and interactions, we introduce $\mathcal{S}$able, versatile model designed comprehensively understand $\mathcal{S}$able incorporates...

At present, the task of identifying crop diseases is mainly to simply distinguish types different diseases. However, current classifiers cannot solve problems, such as accurate identification similar disease categories. Compared with convolutional neural network (CNN), recent vision transformer (VIT) has achieved good results on image tasks. Inspired by this, this paper proposed a multi-granularity feature extraction model based transformer. By combining block information scales, can learn...

In this paper, we study the problem of modeling users' diverse interests. Previous methods usually learn a fixed user representation, which has limited ability to represent distinct interests user. order model various interests, propose Memory Attention-aware Recommender System (MARS). MARS utilizes memory component and novel attentional mechanism deep adaptive representations. Trained in an end-to-end fashion, adaptively summarizes experiments, outperforms seven state-of-the-art on three...

To build a high-quality open-domain chatbot, we introduce the effective training process of PLATO-2 via curriculum learning. There are two stages involved in learning process. In first stage, coarse-grained generation model is trained to learn response under simplified framework one-to-one mapping. second fine-grained generative augmented with latent variables and an evaluation further generate diverse responses select best response, respectively. was on both Chinese English data, whose...

Mood disorders are common and associated with significant morbidity mortality. Early diagnosis has the potential to greatly alleviate burden of mental illness ever increasing costs families society. Mobile devices provide us a promising opportunity detect users' mood in an unobtrusive manner. In this study, we use custom keyboard which collects keystrokes' meta-data accelerometer values. Based on collected time series data multiple modalities, propose deep personalized prediction approach,...