Continual learning for end-to-end automatic speech recognition has to contend with a number of difficulties. Fine-tuning strategies tend to lose performance on data that has been previously trained on, a phenomenon known as catastrophic forgetting. Adapters can help by allowing easy switching between fine-tuned models, but adapted models lose performance on data from other domains, which is a problem if you don’t know what domain your input data comes from. We propose a solution that reduces forgetting to only 3.4% while exceeding the average performance of solutions fine-tuned on all available data, which even with LoRA has a forgetting rate of 49%. Our experiments on diverse datasets and models show that a linear interpolation of several models’ parameters, each fine-tuned from the same generalist model, results in a unified model that performs well on all tested data. In addition, the same model can be fine-tuned and averaged multiple times while maintaining low rates of forgetting.

The non-stationary nature of data streams strongly challenges traditional machine learning techniques. Although some solutions have been proposed to extend traditional machine learning techniques for handling data streams, these approaches either require an initial label set or rely on specialized design parameters. The overlap among classes and the labeling of data streams constitute other major challenges for classifying data streams. In this paper, we proposed a clustering-based data stream classification framework to handle non-stationary data streams without utilizing an initial label set. A density-based stream clustering procedure is used to capture novel concepts with a dynamic threshold and an effective active label querying strategy is introduced to continuously learn the new concepts from the data streams. The sub-cluster structure of each cluster is explored to handle the overlap among classes. Experimental results and quantitative comparison studies reveal that the proposed method provides statistically better or comparable performance than the existing methods.

Standard Convolutional Neural Network (CNN) designs rarely focus on the importance of explicitly capturing diverse features to enhance the network’s performance. Instead, most existing methods follow an indirect approach of increasing or tuning the networks’ depth and width, which in many cases significantly increase the computational cost. Inspired by biological visual system, we proposes a Diverse and Adaptive Attention Convolutional Network (DA2-Net), which enables any feed-forward CNNs to explicitly capture diverse features and adaptively select and emphasize the most informative features to efficiently boost the network’s performance. DA2-Net incurs negligible computational overhead and it is designed to be easily integrated with any CNN architecture. We extensively evaluated DA2-Net on benchmark datasets, including CIFAR100, SVHN, and ImageNet, with various CNN architectures. The experimental results show DA2-Net provides a significant performance improvement with very minimal computational overhead.

In this paper, we present a Robust Completed Local Binary Pattern (RCLBP) framework for a surface defect detection task. Our approach uses a combination of Non-Local (NL) means filter with wavelet thresholding and Completed Local Binary Pattern (CLBP) to extract robust features which are fed into classifiers for surface defects detection. This paper combines three components: A denoising technique based on Non-Local (NL) means filter with wavelet thresholding is established to denoise the noisy image while preserving the textures and edges. Second, discriminative features are extracted using the CLBP technique. Finally, the discriminative features are fed into the classifiers to build the detection model and evaluate the performance of the proposed framework. The performance of the defect detection models are evaluated using a real-world steel surface defect database from Northeastern University (NEU). Experimental results demonstrate that the proposed approach RCLBP is noise robust and can be applied for surface defect detection under varying conditions of intraclass and inter-class changes and with illumination changes.

The growth in the transportation sector has led to an enormous increase in the number of vehicles that ply our roads daily. Even though this advancement has provided numerous transportation modes, it has resulted in serious transportation issues including road congestion. Hence, estimating the number of vehicles on a road will enable traffic managers to take appropriate decisions to curb congestion. In this paper, we propose to use an extreme gradient boosting (XGBoost) algorithm to efficiently and accurately predict the hourly traffic volume. We investigate the effectiveness of the proposed method for different scenarios including how well it performs during extreme weather conditions and holidays. We further investigate the effect of ridge and LASSO regularization on the performance of XGBoost. We then propose a new approach for setting the LASSO regularization parameter in terms of the number of observations and predictors. The performance and computational efficiency of the proposed approach is evaluated on data collected from Interstate-94, Minnesota and the results are compared with existing methods. The results show that the proposed method provides a good balance between performance and computational efficiency.

In this paper, a driver's intention prediction near a road intersection is proposed. Our approach uses a deep bidirectional Long Short-Term Memory (LSTM) with an attention mechanism model based on a hybrid-state system (HSS) framework. As intersection is considered to be as one of the major source of road accidents, predicting a driver's intention at an intersection is very crucial. Our method uses a sequence to sequence modeling with an attention mechanism to effectively exploit temporal information out of the time-series vehicular data including velocity and yaw-rate. The model then predicts ahead of time whether the target vehicle/driver will go straight, stop, or take right or left turn. The performance of the proposed approach is evaluated on a naturalistic driving dataset and results show that our method achieves high accuracy as well as outperforms other methods. The proposed solution is promising to be applied in advanced driver assistance systems (ADAS) and as part of active safety system of autonomous vehicles.

Despite advancements in vehicle security systems, over the last decade, auto-theft rates have increased, and cyber-security attacks on internet-connected and autonomous vehicles are becoming a new threat. In this paper, a deep learning model is proposed, which can identify drivers from their driving behaviors based on vehicle telematics data. The proposed Long-Short-Term-Memory (LSTM) model predicts the identity of the driver based on the individual's unique driving patterns learned from the vehicle telematics data. Given the telematics is time-series data, the problem is formulated as a time series prediction task to exploit the embedded sequential information. The performance of the proposed approach is evaluated on three naturalistic driving datasets, which gives high accuracy prediction results. The robustness of the model on noisy and anomalous data that is usually caused by sensor defects or environmental factors is also investigated. Results show that the proposed model prediction accuracy remains satisfactory and outperforms the other approaches despite the extent of anomalies and noise-induced in the data.

As an effective dimensionality reduction technique, feature selection is widely used in the preprocessing procedure in data mining. It is highly advocated by its superiority in mitigating the effect of noisy data and simplifying the analysis of high-dimensional data. In this paper, a novel unsupervised feature selection procedure based on a clustering algorithm is proposed to evaluate the goodness of features and select a set of useful features without losing the characteristics of the data. It consists of two steps: clustering and feature evaluation. In the clustering procedure, a novel clustering algorithm based on the fitness proportionate sharing is adopted to separate data into distinct clusters without any prior knowledge about data, which is more applicable to the analysis of unknown datasets. On the other hand, the feature evaluation procedure will use the information extracted from the clustering procedure to evaluate the usefulness of each feature and select good features. The proposed method is simulated with four other famous existing feature selection algorithms and a comparison is provided in this paper. Simulation results on both synthetic and real datasets demonstrate that the proposed procedure of feature selection can effectively evaluate the significance of features and obtain a better subset of features than other four existing algorithms.