Comparing feedforward and recurrent neural network architectures with human behavior in artificial grammar learning

A multilayered bidirectional associative memory model for learning nonlinear tasks

A multilayered bidirectional associative memory neural network is proposed to account for learning nonlinear types of association. The model (denoted as the MF-BAM) is composed of two modules, the Multi-Feature extracting bidirectional associative memory (MF), which contains various unsupervised network layers, and a modified Bidirectional Associative Memory (BAM), which consists of a single supervised network layer. The MF generates successive feature patterns from the original inputs. These patterns change the relationship between the inputs and targets in a way that the BAM can learn. The model was tested on different nonlinear tasks, such as the N-bit, Double Moon and its variants, and the 3-class spiral task. Behaviors were reported through learning errors, decision zones, and recall performances. Results showed that it was possible to learn all tasks consistently. By manipulating the number of units per layer and the number of unsupervised network layers in the MF, it was possible to change the level of nonlinearity observed in the decision boundaries. Furthermore, results indicated that different behaviors were achieved from the same set of inputs by using the different generated patterns. These findings are significant as they showed how a BAM-inspired model could solve nonlinear tasks in a more cognitively plausible fashion.

Critical dimension prediction of metal oxide nanoparticle photoresists for electron beam lithography using a recurrent neural network

The critical dimension (CD) of lithographic patterns is the most significant indicator for evaluating the imaging performance of photoresists, and its value is seriously affected by process conditions. However, the lithographic imaging system is highly nonlinear, and extensive exposure experiments are needed to obtain the desired CD. This consumes lots of time, manpower, and cost in screening for optimal process conditions. Here, we report a combined electron beam lithography (EBL) experiment and recurrent neural network (RNN) study on the CDs of metal oxide nanoparticle photoresists, and establish a CD RNN model. Leveraging the RNN model, a process condition filter is developed to screen suitable process conditions. The experimental results demonstrate that the prediction accuracy of the CD model exceeds 93%, and the photoresist patterns under the screened process conditions can satisfy the requirements of a preset CD. This work opens up a novel perspective for accurate EBL process modeling, and provides guidance for EBL experiments.

Recurrent neural network based high-precision position compensation control of magnetic levitation system

For improving the dynamic quality and steady-state performance, the hybrid controller based on recurrent neural network (RNN) is designed to implement the position control of the magnetic levitation ball system in this study. This hybrid controller consists of a baseline controller, an RNN identifier, and an RNN controller. In the hybrid controller, the baseline controller based on the control law of proportional-integral-derivative is firstly employed to provide the online learning sample and maintain the system stability at the early control phase. Then, the RNN identifier is trained online to learn the accurate inverse model of the controlled object. Next, the RNN controller shared the same structures and parameters with the RNN identifier is applied to add the precise compensation control quantity in real-time. Finally, the effectiveness and advancement of the proposed hybrid control strategy are comprehensively validated by the simulation and experimental tests of tracking step, square, sinusoidal, and trapezoidal signals. The results indicate that the RNN-based hybrid controller can obtain higher precision and faster adjustment than the comparison controllers and has strong anti-interference ability and robustness.