Text-to-SQL is the task of translating a natural language query into a structured query language. Existing text-to-SQL approaches focus on improving the model’s architecture while ignoring the relationship between qu...Text-to-SQL is the task of translating a natural language query into a structured query language. Existing text-to-SQL approaches focus on improving the model’s architecture while ignoring the relationship between queries and table schemas and the differences in difficulty between examples in the dataset. To tackle these challenges, a two-stage curriculum learning framework for text-to-SQL(TSCL-SQL) is proposed in this paper. To exploit the relationship between the queries and the table schemas, a schema identification pre-training task is proposed to make the model choose the correct table schema from a set of candidates for a specific query. To leverage the differences in difficulty between examples, curriculum learning is applied to the text-to-SQL task, accompanied by an automatic curriculum learning solution, including a difficulty scorer and a training scheduler. Experiments show that the framework proposed in this paper is effective.展开更多
The Gleason grade group(GG)is an important basis for assessing the malignancy of prostate can-cer,but it requires invasive biopsy to obtain pathology.To noninvasively evaluate GG,an automatic prediction method is prop...The Gleason grade group(GG)is an important basis for assessing the malignancy of prostate can-cer,but it requires invasive biopsy to obtain pathology.To noninvasively evaluate GG,an automatic prediction method is proposed based on multi-scale convolutional neural network of the ensemble attention module trained with curriculum learning.First,a lesion-attention map based on the image of the region of interest is proposed in combination with the bottleneck attention module to make the network more focus on the lesion area.Second,the feature pyramid network is combined to make the network better learn the multi-scale information of the lesion area.Finally,in the network training,a curriculum based on the consistency gap between the visual evaluation and the pathological grade is proposed,which further improves the prediction performance of the network.Ex-perimental results show that the proposed method is better than the traditional network model in predicting GG performance.The quadratic weighted Kappa is 0.4711 and the positive predictive value for predicting clinically significant cancer is 0.9369.展开更多
By leveraging data from a fully labeled source domain,unsupervised domain adaptation(UDA)im-proves classification performance on an unlabeled target domain through explicit discrepancy minimization of data distributio...By leveraging data from a fully labeled source domain,unsupervised domain adaptation(UDA)im-proves classification performance on an unlabeled target domain through explicit discrepancy minimization of data distribution or adversarial learning.As an enhancement,category alignment is involved during adaptation to reinforce target feature discrimination by utilizing model prediction.However,there remain unexplored prob-lems about pseudo-label inaccuracy incurred by wrong category predictions on target domain,and distribution deviation caused by overfitting on source domain.In this paper,we propose a model-agnostic two-stage learning framework,which greatly reduces flawed model predictions using soft pseudo-label strategy and avoids overfitting on source domain with a curriculum learning strategy.Theoretically,it successfully decreases the combined risk in the upper bound of expected error on the target domain.In the first stage,we train a model with distribution alignment-based UDA method to obtain soft semantic label on target domain with rather high confidence.To avoid overfitting on source domain,in the second stage,we propose a curriculum learning strategy to adaptively control the weighting between losses from the two domains so that the focus of the training stage is gradually shifted from source distribution to target distribution with prediction confidence boosted on the target domain.Extensive experiments on two well-known benchmark datasets validate the universal effectiveness of our proposed framework on promoting the performance of the top-ranked UDA algorithms and demonstrate its consistent su-perior performance.展开更多
This study provides a systematic analysis of the resource-consuming training of deep reinforcement-learning (DRL) agents for simulated low-speed automated driving (AD). In Unity, this study established two case studie...This study provides a systematic analysis of the resource-consuming training of deep reinforcement-learning (DRL) agents for simulated low-speed automated driving (AD). In Unity, this study established two case studies: garage parking and navigating an obstacle-dense area. Our analysis involves training a path-planning agent with real-time-only sensor information. This study addresses research questions insufficiently covered in the literature, exploring curriculum learning (CL), agent generalization (knowledge transfer), computation distribution (CPU vs. GPU), and mapless navigation. CL proved necessary for the garage scenario and beneficial for obstacle avoidance. It involved adjustments at different stages, including terminal conditions, environment complexity, and reward function hyperparameters, guided by their evolution in multiple training attempts. Fine-tuning the simulation tick and decision period parameters was crucial for effective training. The abstraction of high-level concepts (e.g., obstacle avoidance) necessitates training the agent in sufficiently complex environments in terms of the number of obstacles. While blogs and forums discuss training machine learning models in Unity, a lack of scientific articles on DRL agents for AD persists. However, since agent development requires considerable training time and difficult procedures, there is a growing need to support such research through scientific means. In addition to our findings, we contribute to the R&D community by providing our environment with open sources.展开更多
Concept learning constructs visual representations that are connected to linguistic semantics, which is fundamental to vision-language tasks. Although promising progress has been made, existing concept learners are st...Concept learning constructs visual representations that are connected to linguistic semantics, which is fundamental to vision-language tasks. Although promising progress has been made, existing concept learners are still vulnerable to attribute perturbations and out-of-distribution compositions during inference. We ascribe the bottleneck to a failure to explore the intrinsic semantic hierarchy of visual concepts, e.g., {red, blue,···} ∈“color” subspace yet cube ∈“shape”. In this paper, we propose a visual superordinate abstraction framework for explicitly modeling semantic-aware visual subspaces(i.e., visual superordinates). With only natural visual question answering data, our model first acquires the semantic hierarchy from a linguistic view and then explores mutually exclusive visual superordinates under the guidance of linguistic hierarchy. In addition, a quasi-center visual concept clustering and superordinate shortcut learning schemes are proposed to enhance the discrimination and independence of concepts within each visual superordinate. Experiments demonstrate the superiority of the proposed framework under diverse settings, which increases the overall answering accuracy relatively by 7.5% for reasoning with perturbations and 15.6% for compositional generalization tests.展开更多
With the emergence of pre-trained models,current neural networks are able to give task performance that is comparable to humans.However,we know little about the fundamental working mechanism of pre-trained models in w...With the emergence of pre-trained models,current neural networks are able to give task performance that is comparable to humans.However,we know little about the fundamental working mechanism of pre-trained models in which we do not know how they approach such performance and how the task is solved by the model.For example,given a task,human learns from easy to hard,whereas the model learns randomly.Undeniably,difficulty-insensitive learning leads to great success in natural language processing(NLP),but little attention has been paid to the effect of text difficulty in NLP.We propose a human learning matching index(HLM Index)to investigate the effect of text difficulty.Experiment results show:1)LSTM gives more human-like learning behavior than BERT.Additionally,UID-SuperLinear gives the best evaluation of text difficulty among four text difficulty criteria.Among nine tasks,some tasks’performance is related to text difficulty,whereas others are not.2)Model trained on easy data performs best in both easy and medium test data,whereas trained on hard data only performs well on hard test data.3)Train the model from easy to hard,leading to quicker convergence.展开更多
基金Fundamental Research Funds for the Central Universities,China (No. 2232023D-19)。
文摘Text-to-SQL is the task of translating a natural language query into a structured query language. Existing text-to-SQL approaches focus on improving the model’s architecture while ignoring the relationship between queries and table schemas and the differences in difficulty between examples in the dataset. To tackle these challenges, a two-stage curriculum learning framework for text-to-SQL(TSCL-SQL) is proposed in this paper. To exploit the relationship between the queries and the table schemas, a schema identification pre-training task is proposed to make the model choose the correct table schema from a set of candidates for a specific query. To leverage the differences in difficulty between examples, curriculum learning is applied to the text-to-SQL task, accompanied by an automatic curriculum learning solution, including a difficulty scorer and a training scheduler. Experiments show that the framework proposed in this paper is effective.
基金Foundation item:the Suzhou Municipal Health and Family Planning Commission's Key Diseases Diagnosis and Treatment Program(No.LCzX202001)the Science and Technology Development Project ofSuzhou(Nos.SS2019012andSKY2021031)+1 种基金the Youth Innovation Promotion Association CAS(No.2021324)the Medical Research Project of Jiangsu Provincial Health and Family Planning Commission(No.M2020068)。
文摘The Gleason grade group(GG)is an important basis for assessing the malignancy of prostate can-cer,but it requires invasive biopsy to obtain pathology.To noninvasively evaluate GG,an automatic prediction method is proposed based on multi-scale convolutional neural network of the ensemble attention module trained with curriculum learning.First,a lesion-attention map based on the image of the region of interest is proposed in combination with the bottleneck attention module to make the network more focus on the lesion area.Second,the feature pyramid network is combined to make the network better learn the multi-scale information of the lesion area.Finally,in the network training,a curriculum based on the consistency gap between the visual evaluation and the pathological grade is proposed,which further improves the prediction performance of the network.Ex-perimental results show that the proposed method is better than the traditional network model in predicting GG performance.The quadratic weighted Kappa is 0.4711 and the positive predictive value for predicting clinically significant cancer is 0.9369.
基金the 111 Project(No.BP0719010)the Project of the Science and Technology Commission of Shanghai Municipality(No.18DZ2270700)。
文摘By leveraging data from a fully labeled source domain,unsupervised domain adaptation(UDA)im-proves classification performance on an unlabeled target domain through explicit discrepancy minimization of data distribution or adversarial learning.As an enhancement,category alignment is involved during adaptation to reinforce target feature discrimination by utilizing model prediction.However,there remain unexplored prob-lems about pseudo-label inaccuracy incurred by wrong category predictions on target domain,and distribution deviation caused by overfitting on source domain.In this paper,we propose a model-agnostic two-stage learning framework,which greatly reduces flawed model predictions using soft pseudo-label strategy and avoids overfitting on source domain with a curriculum learning strategy.Theoretically,it successfully decreases the combined risk in the upper bound of expected error on the target domain.In the first stage,we train a model with distribution alignment-based UDA method to obtain soft semantic label on target domain with rather high confidence.To avoid overfitting on source domain,in the second stage,we propose a curriculum learning strategy to adaptively control the weighting between losses from the two domains so that the focus of the training stage is gradually shifted from source distribution to target distribution with prediction confidence boosted on the target domain.Extensive experiments on two well-known benchmark datasets validate the universal effectiveness of our proposed framework on promoting the performance of the top-ranked UDA algorithms and demonstrate its consistent su-perior performance.
文摘This study provides a systematic analysis of the resource-consuming training of deep reinforcement-learning (DRL) agents for simulated low-speed automated driving (AD). In Unity, this study established two case studies: garage parking and navigating an obstacle-dense area. Our analysis involves training a path-planning agent with real-time-only sensor information. This study addresses research questions insufficiently covered in the literature, exploring curriculum learning (CL), agent generalization (knowledge transfer), computation distribution (CPU vs. GPU), and mapless navigation. CL proved necessary for the garage scenario and beneficial for obstacle avoidance. It involved adjustments at different stages, including terminal conditions, environment complexity, and reward function hyperparameters, guided by their evolution in multiple training attempts. Fine-tuning the simulation tick and decision period parameters was crucial for effective training. The abstraction of high-level concepts (e.g., obstacle avoidance) necessitates training the agent in sufficiently complex environments in terms of the number of obstacles. While blogs and forums discuss training machine learning models in Unity, a lack of scientific articles on DRL agents for AD persists. However, since agent development requires considerable training time and difficult procedures, there is a growing need to support such research through scientific means. In addition to our findings, we contribute to the R&D community by providing our environment with open sources.
基金supported in part by the Australian Research Council(ARC)(Nos.FL-170100117,DP-180103424,IC-190100031 and LE-200100049).
文摘Concept learning constructs visual representations that are connected to linguistic semantics, which is fundamental to vision-language tasks. Although promising progress has been made, existing concept learners are still vulnerable to attribute perturbations and out-of-distribution compositions during inference. We ascribe the bottleneck to a failure to explore the intrinsic semantic hierarchy of visual concepts, e.g., {red, blue,···} ∈“color” subspace yet cube ∈“shape”. In this paper, we propose a visual superordinate abstraction framework for explicitly modeling semantic-aware visual subspaces(i.e., visual superordinates). With only natural visual question answering data, our model first acquires the semantic hierarchy from a linguistic view and then explores mutually exclusive visual superordinates under the guidance of linguistic hierarchy. In addition, a quasi-center visual concept clustering and superordinate shortcut learning schemes are proposed to enhance the discrimination and independence of concepts within each visual superordinate. Experiments demonstrate the superiority of the proposed framework under diverse settings, which increases the overall answering accuracy relatively by 7.5% for reasoning with perturbations and 15.6% for compositional generalization tests.
基金the support of the National Natural Science Foundation of China(Nos.U22B2059,62176079)National Natural Science Foundation of Heilongjiang Province,China(No.YQ 2022F005)the Industry-University-Research Innovation Foundation of China University(No.2021ITA05009).
文摘With the emergence of pre-trained models,current neural networks are able to give task performance that is comparable to humans.However,we know little about the fundamental working mechanism of pre-trained models in which we do not know how they approach such performance and how the task is solved by the model.For example,given a task,human learns from easy to hard,whereas the model learns randomly.Undeniably,difficulty-insensitive learning leads to great success in natural language processing(NLP),but little attention has been paid to the effect of text difficulty in NLP.We propose a human learning matching index(HLM Index)to investigate the effect of text difficulty.Experiment results show:1)LSTM gives more human-like learning behavior than BERT.Additionally,UID-SuperLinear gives the best evaluation of text difficulty among four text difficulty criteria.Among nine tasks,some tasks’performance is related to text difficulty,whereas others are not.2)Model trained on easy data performs best in both easy and medium test data,whereas trained on hard data only performs well on hard test data.3)Train the model from easy to hard,leading to quicker convergence.