Domain Specific Learning for Sentiment Classification and Activity Recognition

Wang, Hongbo; Xue, Yanze; Zhen, Xiaoxiao; Tu, Xuyan

doi:10.1109/access.2018.2871349

Cited by 8 publications

(3 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Ma M et.al [33] proposed a twin stream network architecture and jointly fine-tuned the two networks to recognize objects, actions, and activities. Wang H B et.al [52] verified that fine-tuning and regular constraints can increase the training efficiency, and fine-tuning is valid in practical HAR applications.…”

Section: Transfer Learning and Domain Adaptationmentioning

confidence: 96%

Domain Generalization for Activity Recognition via Adaptive Feature Fusion

Xin¹,

Wang²,

Chen³

et al. 2022

Preprint

View full text Add to dashboard Cite

Human activity recognition requires the efforts to build a generalizable model using the training datasets with the hope to achieve good performance in test datasets. However, in real applications, the training and testing datasets may have totally different distributions due to various reasons such as different body shapes, acting styles, and habits, damaging the model's generalization performance. While such a distribution gap can be reduced by existing domain adaptation approaches, they typically assume that the test data can be accessed in the training stage, which is not realistic. In this paper, we consider a more practical and challenging scenario: domain-generalized activity recognition (DGAR) where the test dataset cannot be accessed during training. To this end, we propose Adaptive Feature Fusion for Activity Recognition (AFFAR), a domain generalization approach that learns to fuse the domain-invariant and domain-specific representations to improve the model's generalization performance. AFFAR takes the best of both worlds where domaininvariant representations enhance the transferability across domains and domain-specific representations leverage the model discrimination power from each domain. Extensive experiments on three public HAR datasets show its effectiveness. Furthermore, we apply AFFAR to a real application, i.e., the diagnosis of Children's Attention Deficit Hyperactivity Disorder (ADHD), which also demonstrates the superiority of our approach.

show abstract

Section: Transfer Learning and Domain Adaptationmentioning

confidence: 96%

Domain Generalization for Activity Recognition via Adaptive Feature Fusion

Xin¹,

Wang²,

Chen³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…This machine is stacked with RBM, which has powerful feature extraction capabilities. The restricted Boltzmann machine [28][29] is a Markov random field model, which has a 2-layer structure, as shown in Figure 1. The lower layer is the input layer, containing input units , used to represent input data, and each input unit contains a real-valued offset ; the upper layer is the hidden layer, containing hidden units ℎ , which represent the input extracted by RBM abstract feature of data, each hidden unit contains a real-valued bias .…”

Section: A Deep Belief Networkmentioning

confidence: 99%

“…In order to highlight the powerful non-linear fitting ability, Since the amount of source training data is relatively large relative to the amount of task data, and the amount of data to be calculated is small, in order to test whether the amount of task data affects the migration result when the source training data is migrated based on the maximum mean difference contribution coefficient method, auxiliary sample data is introduced and set are 10 auxiliary sample batches, and the data of August 22, August (22)(23), August (22)(23)(24),..., August (22)(23)(24)(25)(26)(27)(28)(29)(30)(31) are taken as 10 sample data, the number of samples is 96,192,...,960 in sequence. Then take the data under different auxiliary samples as the target data, migrate data close to the target data distribution from the source data, calculate the MMD value of each auxiliary sample, the source data, and the migrated data respectively, and use the migration data of each auxiliary sample The TDBN-DNN model obtained after finetuning the network calculates the target task data, and the Gaussian kernel width control parameter = 2.…”

Section: Comparison Of Dbn-dnn and Tdbn-dnn Algorithmsmentioning

confidence: 99%