Learning Accelerators Research Articles

Precision teaching in the 21st century: trends, barriers and the path forward

Purpose This paper aims to explore the current state of Precision Teaching, identifies key barriers the field must address and outlines future directions for progress. Design/methodology/approach Precision Teaching (PT) is a methodology that uses the family of Standard Celeration Charts to monitor changes in behaviour frequencies, guide instructional decisions and promote the acceleration of learning. Since its inception, PT has been widely adopted across general and special education and other settings. Research has demonstrated its effectiveness in improving skill acquisition in interventions, including reading fluency, mathematical skills and critical thinking. Findings Despite its success, barriers to broader implementation remain, including limited awareness, technological challenges and research promotion. As the field continues to evolve, interdisciplinary collaboration and technology promotion will be key to ensuring the continued growth of PT. Originality/value By addressing these barriers, the methodology can expand further, enhancing its impact across a diverse array of fields and populations.

Akselerasi Pembelajaran Tahfiz Pada Lembaga Pendidikan Al Quran Di Fath Darut Tafsir Sukawangi Bogor

This research is motivated by the author's interest in three things that are problems regarding learning tahfiz Al-Quran. First, an Al-Quran educational institution with an Al-Quran tahfiz program must prepare professional policies in tahfiz learning strategies and infrastructure so that the goals of Al-Quran tahfiz are achieved as determined. Second, the process of learning tahfiz Al-Quran really requires the right approach, methods and evaluation, so that learning runs well and smoothly without any significant obstacles in achieving success. Third, accelerated learning of Al-Quran tahfiz is really needed by students because there are many branches of knowledge that need to be studied afterwards, both religious knowledge (such as tafsir, fiqh, etc.) or general knowledge. This research aims to (1) analyze the effectiveness of Islamic boarding school leaders' strategies and policies in accelerating the learning of Al-Quran tahfiz. (2) Examining strategies for learning Al-Quran tahfiz at Islamic boarding schools. (3) Finding the main success factors in accelerating learning of Al-Quran tahfiz in Islamic boarding schools.In connection with the problem raised by the author, namely the acceleration of tahfiz learning at Al-Quran educational institutions at the Fath Darut Tafsir Islamic Boarding School, the author uses qualitative research methods at the Fath Darut Tafsir Islamic Boarding School to encourage discipline among students, so that the UTQ Tahfidz Qur'an acceleration program ('Uzlah Tahfidz Qur'an) at the Fath Darut Tafsir Islamic Boarding School runs smoothly. This program was attended by all Islamic boarding school elements from santri & female students, totaling around 150 people, 17 Ustadz & Ustadzah as Musyrif or Musyrifah halaqah of the Qur'an. In the UTQ program, 17 halaqahs will be made for 150 santri and female students with a target of 10 days to get 5 new Juz memorized, and it is hoped that if the students have participated in this UTQ event 6 times they will be able to complete memorizing 30 Juz of the Qur'an.As a result of the UTQ Program, a total of 30 students managed to memorize 5 new juz with a percentage of 20% of the students who took part in the accelerated program. Meanwhile, 15 santri and santriwati managed to memorize more than 5 juz if the percentage was 10%, and those who managed to memorize 4 juz were 40% and less than 4 juz were 30%.

A Real-Time Signal Measurement System Using FPGA-Based Deep Learning Accelerators and Microwave Photonic

A Real-Time Signal Measurement System Using FPGA-Based Deep Learning Accelerators and Microwave Photonic

Method for Detecting Roadbed Compaction Degree Based on Machine Learning and Vibration Acceleration

Roadbed construction occupies a core position in highway construction, but its quality is easily constrained by multiple factors such as changing environmental factors, the performance of construction equipment, and the professional abilities of construction personnel, leading to potential quality risks. Traditional quality inspection methods are mostly carried out after construction is completed, making it difficult to achieve continuous and real-time monitoring of roadbed compaction quality, which to some extent limits the real-time feedback and adjustment of construction quality. Vibration compaction technology has been widely used in the field of highway engineering due to its high efficiency and speed. The compaction degree is directly related to the durability and service life of the highway; therefore, accurate and efficient detection of compaction degree is crucial. This article proposes a method for detecting roadbed compaction degree by integrating machine learning (ML) and vibration acceleration signals. This method aims to achieve accurate evaluation of roadbed compaction by real-time monitoring and analysis of vibration acceleration data, combined with the powerful prediction and classification capabilities of ML algorithms. The experimental results show that this method not only improves the detection efficiency, but also significantly enhances the accuracy of compaction degree detection.

Beam Orbital Parameter Prediction Based on the Deployment of Cascaded Neural Networks at Edge Intelligence Acceleration Nodes

During the beam current calibration process, accurate guidance of the beam current to the metal target is a challenging issue for proton accelerators. To address this challenge, we propose the use of beam orbital parameters combined with reinforcement learning algorithms to achieve automatic beam calibration. This study introduces a system architecture that employs edge intelligent acceleration nodes based on deep learning acceleration techniques. We designed a system to predict BPM parameters using a cascaded backpropagation neural network (CBPNN) that is informed by the physical structure. This system serves as an environmental map for reinforcement learning, aiding beam current correction. The CBPNN was implemented on the acceleration node to hasten the forward inference process, leveraging sparsification, quantization algorithms, and pipelining techniques. Our experimental results demonstrated that the simulated inference speed reached 28 μs with FPGA hardware as the edge acceleration node, achieving forward inference speeds 35.66 and 12.66 times faster than those of the CPU and GPU. The energy efficiency ratio was 10.582 MOPS/W, which was 989 and 410 times that of the CPU and GPU, respectively. This confirms the designed architecture’s energy efficiency and low latency attributes.

English First Additional Language teachers’ perceptions of translanguaging pedagogy: A case of Vhembe District, Limpopo Province

This article reports on Grade 7 English First Additional Language (FAL) teachers’ perceptions of their use of a translanguaging pedagogy in teaching English FAL among Grade 7 learners in Vhembe District, Limpopo Province. The Continua of biliteracy model (CBM) underpinned the study. CBM foregrounds the development of biliteracy among bi/multilinguals and supports the view that learners’ communicative repertoires and language practices form an integral part of their learning process. The participants in the study were 6 Grade 7 English FAL teachers purposefully selected from three different primary schools. The study collected data through classroom observations, open-ended questionnaires, and semi-structured interviews. The overall results of the study revealed two differentiated strands of the teachers’ perceptions of the translanguaging approach in the teaching and learning of English FAL among Grade 7 learners. Firstly, the majority of Grade 7 teachers have a positive outlook on translanguaging pedagogy in the teaching and learning of English FAL. Alternation and flexible use of Tshivenḓa HL and English are permitted for scaffolding and acceleration of L2 learning. Secondly, the study revealed that some of these Grade 7 English FAL teachers objected to the use of translanguaging in preference for an English-only approach to accelerate the learning of English in their L2 classrooms. In their view, Tshivenḓa Home Language and English must be taught in isolation because of their distinct linguistic systems.

Efficient SpMM Accelerator for Deep Learning: Sparkle and Its Automated Generator

Deep learning (DL) technology has made breakthroughs in a wide range of intelligent tasks, such as vision, language, recommendation systems, and so on. Sparse matrix multiplication (SpMM) is the key computation kernel of most sparse models. Conventional computing platforms, such as CPUs, GPUs, and AI chips with regular processing units, are unable to effectively support sparse computation due to their fixed structure and instruction sets. This work extends Sparkle, an accelerator architecture, which is developed specifically for processing SpMM in DL. During the balanced data loading process, some modifications are implemented to enhance the flexibility of the Sparkle architecture. Additionally, a Sparkle generator is proposed to accommodate diverse resource constraints and facilitate adaptable deployment. Leveraging Sparkle’s structural parameters and template-based design methods, the generator enables automatic Sparkle circuit generation under varying parameters. An instantiated Sparkle accelerator is implemented on the Xilinx xqvu11p FPGA platform with a specific configuration. Compared to the state-of-the-art SpMM accelerator SIGMA, the Sparkle accelerator instance improves the sparse computing efficiency by about 10 to 20 \(\%\) . Furthermore, the Sparkle instance achieved 7.76 \(\times\) higher performance over the Nvidia Orin NX GPU. More instances of accelerators with different parameters were evaluated, demonstrating that the Sparkle architecture can effectively accelerate SpMM.

N-AquaRAM: A Cost-Efficient Deep Learning Accelerator for Real-Time Aquaponic Monitoring

AbstractAquaponics is an emerging area of agricultural sciences that combines aquaculture and hydroponics in a symbiotic way to increase crop production. Though it offers a lot of advantages over traditional techniques, including chemical-free and soil-less farming, its commercial application suffers from some problems such as the lack of experienced manpower. To operate a stable smart aquaponic system, it is critical to estimate the fish size properly. In this context, the use of dedicated hardware for real-time aquaponic monitoring can greatly resolve the issue of inexperienced handlers. In this article, we present a complete methodology to train a deep neural network to perform fish size estimation in real time. To achieve high accuracy, a novel implementation of swish function is presented. This novel version is far more hardware efficient than the original one, while being extremely accurate. Moreover, we present a deep learning accelerator that can classify 40 million fish samples in a second. The dedicated real-time system is about 1600 times faster than the one based on general-purpose computers. The proposed neuromorphic accelerator consumes about 2600 slice registers on a low-end model of Virtex 6 FPGA series.

T-phPINN: Physics-informed neural networks for solving 2D non-Fourier heat conduction equations

Non-Fourier heat conduction plays a dominant role in many extreme transient heat conduction processes, such as laser pulses and heat transfer in biological systems, but the heat wave effect makes it difficult to solve the temperature field accurately and quickly. In order to solve this problem, the first order time derivative enhanced parallel hard constraints physics-informed neural networks (T-phPINN) is proposed. T-phPINN comprises two subnetworks and incorporates a first order time derivative to capture sharp temperature changes. Two numerical cases show that the minimum relative error of T-phPINN is 0.001 % and 0.015 %, which is 1.04 % and 12.30 % of the error of conventional PINN respectively, proving the accuracy of our architecture. A transfer learning framework is established for scenarios of different parameters, the training only requires 1/6 iterations of the basic model, and close accuracy is obtained. The computational cost of T-phPINN is evaluated using the finite element method as the baseline. For the two cases, the single calculation time is 33.43 % and 51.50 % of the baseline, while the multiple calculation time under the acceleration of transfer learning is 11.59 % and 17.75 % of the baseline. This study will be helpful for solving large-scale non-Fourier heat conduction equations precisely and expeditiously.

DLAS: A Conceptual Model for Across-Stack Deep Learning Acceleration

Deep Neural Networks (DNNs) are very computationally demanding, which presents a significant barrier to their deployment, especially on resource-constrained devices. Significant work from both the machine learning and computing systems communities has attempted to accelerate DNNs. However, the number of techniques available and the required domain knowledge for their exploration continues to grow, making design space exploration (DSE) increasingly difficult. To unify the perspectives from these two communities, this paper introduces the Deep Learning Acceleration Stack (DLAS), a conceptual model for DNN deployment and acceleration. We adopt a six-layer representation that organizes and illustrates the key areas for DNN acceleration, from machine learning to software and computer architecture. We argue that the DLAS model balances simplicity and expressiveness, assisting practitioners from various domains in tackling co-design acceleration challenges. We demonstrate the interdependence of the DLAS layers, and thus the need for co-design, through an across-stack perturbation study, using a modified tensor compiler to generate experiments for combinations of a few parameters across the DLAS layers. Our perturbation study assesses the impact on inference time and accuracy when varying DLAS parameters across two datasets, seven popular DNN architectures, four compression techniques, three algorithmic primitives (with sparse and dense variants), untuned and auto-scheduled code generation, and four hardware platforms. The study observes significant changes in the relative performance of design choices with the introduction of new DLAS parameters (e.g., the fastest algorithmic primitive varies with the level of quantization). Given the strong evidence for the need for co-design, and the high costs of DSE, DLAS offers a valuable conceptual model for better exploring advanced co-designed accelerated deep learning solutions.

Learning to Guide Particle Search for Dynamic Multiobjective Optimization.

Dynamic multiobjective optimization problems (DMOPs) are characterized by multiple objectives that change over time in varying environments. More specifically, environmental changes can be described as various dynamics. However, it is difficult for existing dynamic multiobjective algorithms (DMOAs) to handle DMOPs due to their inability to learn in different environments to guide the search. Besides, solving DMOPs is typically an online task, requiring low computational cost of a DMOA. To address the above challenges, we propose a particle search guidance network (PSGN), capable of directing individuals' search actions, including learning target selection and acceleration coefficient control. PSGN can learn the actions that should be taken in each environment through rewarding or punishing the network by reinforcement learning. Thus, PSGN is capable of tackling DMOPs of various dynamics. Additionally, we efficiently adjust PSGN hidden nodes and update the output weights in an incremental learning way, enabling PSGN to direct particle search at a low computational cost. We compare the proposed PSGN with seven state-of-the-art algorithms, and the excellent performance of PSGN verifies that it can handle DMOPs of various dynamics in a computationally very efficient way.

Sense of Embodiment Supports Motor Learning and Adaptation in Tele-robotics

In this paper, we transition from the theoretical and experimental groundwork of manipulating and measuring Sense of Embodiment (SoE) to addressing a fundamental question: What is the purpose of optimizing the SoE in a teleoperation system? This exploration centers on investigating the potential positive effects of SoE on motor adaptation, the acceleration of motor learning, and the potential enhancement of task performance. The paper delves into this investigation by focusing on two critical research questions: 1) what is the effect of SoE on task performance in a perceptual-motor task? 2) What is the effect of SoE on the asymptote of the learning curve in a perceptual-motor task? Drawing insights from the existing literature, the hypothesis emerges that enhancing SoE yields positive effects, not only on task performance (H1) but also on the overall embodiment experience (H2). An additional layer of exploration is introduced through an exploratory research question: Are these results consistent across diverse scenarios and tasks? The study design encompasses two distinct user studies, each set in different applications and featuring various avatars, yet all anchored in similar tasks, specifically a modified peg-in-hole task: 1) in the first experiment, participants operated a robotic arm with a human-like hand as end-effector, and they were required to perform a classic peg-in-hole task; 2) in user study 2, the task is transformed into a variation that we called “peg-on-button”, wherein participants use a robotic arm with a gripper as the end-effector to press a lit button. In both studies, a consistent pattern emerges: a setup that fosters embodiment has a positive impact on motor learning and adaptation, resulting in improved task performance. A supportive setup also reduces the perception of the surrogate as a mere mediator between the operator and the remote environment, especially when contrasted with a setup that suppresses embodiment. The positive effects on motor learning and task performance advocate for the incorporation of embodiment-supportive designs in teleoperational setups. However, the nuanced relationship between SoE and long-term task performance prompts a call for further exploration and consideration of various factors influencing teleoperation outcomes across diverse scenarios and tasks.

Symmetric silicon microring resonator optical crossbar array for accelerated inference and training in deep learning

Photonic integrated circuits are emerging as a promising platform for accelerating matrix multiplications in deep learning, leveraging the inherent parallel nature of light. Although various schemes have been proposed and demonstrated to realize such photonic matrix accelerators, the in situ training of artificial neural networks using photonic accelerators remains challenging due to the difficulty of direct on-chip backpropagation on a photonic chip. In this work, we propose a silicon microring resonator (MRR) optical crossbar array with a symmetric structure that allows for simple on-chip backpropagation, potentially enabling the acceleration of both the inference and training phases of deep learning. We demonstrate a 4×4 circuit on a Si-on-insulator platform and use it to perform inference tasks of a simple neural network for classifying iris flowers, achieving a classification accuracy of 93.3%. Subsequently, we train the neural network using simulated on-chip backpropagation and achieve an accuracy of 91.1% in the same inference task after training. Furthermore, we simulate a convolutional neural network for handwritten digit recognition, using a 9×9 MRR crossbar array to perform the convolution operations. This work contributes to the realization of compact and energy-efficient photonic accelerators for deep learning.

Transcutaneous cervical vagus nerve stimulation enhances second-language vocabulary acquisition while simultaneously mitigating fatigue and promoting focus

Transcutaneous vagus nerve stimulation (tVNS) is a promising technique for enhancing cognitive performance and skill acquisition. Yet, its efficacy for enhancing learning rate and long-term retention in an ecologically valid learning environment has not been demonstrated. We conducted two double-blind sham-controlled experiments examining the efficacy of auricular tVNS (taVNS: Experiment (1) and cervical tVNS (tcVNS: Experiment (2), on a 5 day second-language vocabulary acquisition protocol among highly selected career linguists at the US Department of Defense’s premier language school. tcVNS produced accelerated recall performance during training (Day 2–4), benefits of which were maintained across a 24 h retention interval with no stimulation at the final test. Consistent with prior work, tcVNS also produced fatigue-mitigating and focus-promoting effects as measured by the Air Force Research Laboratory Mood Questionnaire. Based on the current and the previous findings supporting tVNS’ efficacy on performance, training enhancement, and fatigue mitigation, we believe tcVNS to be an effective learning acceleration tool that can be utilized at language-teaching and other institutions focused on intensive training of cognitive skills.

Low-power, high-speed comparator design at 45-nm CMOS for efficient deep learning acceleration

Low-power, high-speed comparator design at 45-nm CMOS for efficient deep learning acceleration

A computational perspective on neural-symbolic integration

Neural-Symbolic Integration (NSI) aims to marry the principles of symbolic AI techniques, such as logical reasoning, with the learning capabilities of neural networks. In recent years, many systems have been proposed to address this integration in a seemingly efficient manner. However, from the computational perspective, this is in principle impossible to do. Specifically, some of the core symbolic problems are provably hard, hence a general NSI system necessarily needs to adopt this computational complexity, too. Many NSI methods try to circumvent this downside by inconspicuously dropping parts of the symbolic capabilities while mapping the problems into static tensor representations in exchange for efficient deep learning acceleration. In this paper, we argue that the aim for a general NSI system, properly covering both the neural and symbolic paradigms, has important computational implications on the learning representations, the structure of the resulting computation graphs, and the underlying hardware and software stacks. Particularly, we explain how the currently prominent, tensor-based deep learning with static computation graphs is conceptually insufficient as a foundation for such general NSI, which we discuss in a wider context of established (statistical) relational and structured deep learning methods. Finally, we delve into the underlying hardware acceleration aspects and outline some promising computational directions toward fully expressive and efficient NSI.

A review of emerging trends in photonic deep learning accelerators

Deep learning has revolutionized many sectors of industry and daily life, but as application scale increases, performing training and inference with large models on massive datasets is increasingly unsustainable on existing hardware. Highly parallelized hardware like Graphics Processing Units (GPUs) are now widely used to improve speed over conventional Central Processing Units (CPUs). However, Complementary Metal-oxide Semiconductor (CMOS) devices suffer from fundamental limitations relying on metallic interconnects which impose inherent constraints on bandwidth, latency, and energy efficiency. Indeed, by 2026, the projected global electricity consumption of data centers fueled by CMOS chips is expected to increase by an amount equivalent to the annual usage of an additional European country. Silicon Photonics (SiPh) devices are emerging as a promising energy-efficient CMOS-compatible alternative to electronic deep learning accelerators, using light to compute as well as communicate. In this review, we examine the prospects of photonic computing as an emerging solution for acceleration in deep learning applications. We present an overview of the photonic computing landscape, then focus in detail on SiPh integrated circuit (PIC) accelerators designed for different neural network models and applications deep learning. We categorize different devices based on their use cases and operating principles to assess relative strengths, present open challenges, and identify new directions for further research.

An Open-Source ML-Based Full-Stack Optimization Framework for Machine Learning Accelerators

Parameterizable machine learning (ML) accelerators are the product of recent breakthroughs in ML. To fully enable their design space exploration (DSE), we propose a physical-design-driven, learning-based prediction framework for hardware-accelerated deep neural network (DNN) and non-DNN ML algorithms. It adopts a unified approach that combines power, performance, and area (PPA) analysis with frontend performance simulation, thereby achieving a realistic estimation of both backend PPA and system metrics such as runtime and energy. In addition, our framework includes a fully automated DSE technique, which optimizes backend and system metrics through an automated search of architectural and backend parameters. Experimental studies show that our approach consistently predicts backend PPA and system metrics with an average 7% or less prediction error for the ASIC implementation of two deep learning accelerator platforms, VTA and VeriGOOD-ML, in both a commercial 12 nm process and a research-oriented 45 nm process.

Factored Systolic Arrays Based on Radix-8 Multiplication for Machine Learning Acceleration

Factored Systolic Arrays Based on Radix-8 Multiplication for Machine Learning Acceleration

Distributed transformer for high order epistasis detection in large-scale datasets

Understanding the genetic basis of complex diseases is one of the most important challenges in current precision medicine. To this end, Genome-Wide Association Studies aim to correlate Single Nucleotide Polymorphisms (SNPs) to the presence or absence of certain traits. However, these studies do not consider interactions between several SNPs, known as epistasis, which explain most genetic diseases. Analyzing SNP combinations to detect epistasis is a major computational task, due to the enormous search space. A possible solution is to employ deep learning strategies for genomic prediction, but the lack of explainability derived from the black-box nature of neural networks is a challenge yet to be addressed. Herein, a novel, flexible, portable, and scalable framework for network interpretation based on transformers is proposed to tackle any-order epistasis. The results on various epistasis scenarios show that the proposed framework outperforms state-of-the-art methods for explainability, while being scalable to large datasets and portable to various deep learning accelerators. The proposed framework is validated on three WTCCC datasets, identifying SNPs related to genes known in the literature that have direct relationships with the studied diseases.