European Journal of Computer Science and Information Technology,13(17),88-98, 2025

Print ISSN: 2054-0957 (Print)

Online ISSN: 2054-0965 (Online)
Website: https://www.eajournals.org/
Publication of the European Centre for Research Training and Development -UK

The Rise of Deep Learning and Neural

Networks: Revolutionizing Artificial
Intelligence
Chirag Gajiwala
Nutanix Inc., USA

doi: https://doi.org/10.37745/ejcsit.2013/vol13n178898 Published May 11, 2025

Citation: Gajiwala C. (2025) The Rise of Deep Learning and Neural Networks: Revolutionizing Artificial
Intelligence, European Journal of Computer Science and Information Technology,13(17),88-99

Abstract: This comprehensive article explores the transformative impact of deep learning and neural
networks on artificial intelligence and various industries. It delves into the fundamental principles of deep
learning, highlighting its remarkable performance in tasks such as image recognition, natural language
processing, and speech recognition. The article examines the widespread adoption of deep learning across
sectors including healthcare, automotive, and NLP, showcasing its potential to revolutionize processes and
unlock new possibilities. It also discusses recent advancements in AI research, particularly in reinforcement
learning and generative models, and looks ahead to future prospects such as improved interpretability,
energy-efficient models, multi-modal learning, and neuromorphic computing. The economic impact and
potential challenges of this rapidly evolving field are also addressed, emphasizing the need for responsible
development and deployment of these technologies.

Keywords: deep learning, neural networks, artificial intelligence, machine learning, industry applications

INTRODUCTION

In recent years, deep learning and neural networks have emerged as transformative technologies within the
field of artificial intelligence (AI), revolutionizing the landscape of machine learning and driving
unprecedented advancements across various industries. The global deep learning market, valued at $10.2
billion in 2022, is projected to reach a staggering $93.8 billion by 2028, growing at a compound annual
growth rate (CAGR) of 39.2% [1]. This exponential growth underscores the increasing adoption and impact
of these advanced techniques in pushing the boundaries of what's possible in AI.

Deep learning, a subset of machine learning inspired by the structure and function of the human brain, has
demonstrated remarkable capabilities in handling complex, high-dimensional data. Neural networks, the

88
 European Journal of Computer Science and Information Technology,13(17),88-98, 2025
Print ISSN: 2054-0957 (Print)
Online ISSN: 2054-0965 (Online)
Website: https://www.eajournals.org/
Publication of the European Centre for Research Training and Development -UK
cornerstone of deep learning, have shown exceptional performance in tasks such as image and speech
recognition, natural language processing, and autonomous decision-making. For instance, state-of-the-art
deep learning models have achieved human-level or superhuman performance in various domains, with
error rates as low as 3.5% in image classification tasks and word error rates of just 5.9% in speech
recognition [2].

The transformative power of deep learning is evident in its wide-ranging applications across industries. In
healthcare, deep learning algorithms are revolutionizing medical imaging diagnostics, with studies showing
accuracy rates of up to 95% in detecting certain types of cancer, potentially saving countless lives through
early detection. The automotive sector has seen a paradigm shift with the development of autonomous
vehicles, where deep neural networks process vast amounts of sensory data in real-time, making split-
second decisions with an accuracy that rivals human drivers.

Moreover, the field of natural language processing (NLP) has experienced a quantum leap, with deep
learning models like transformers enhancing language translation, sentiment analysis, and chatbots. These
advancements have led to machine translation systems that can handle over 100 languages with near-human
accuracy, and chatbots capable of engaging in nuanced, context-aware conversations across a multitude of
domains.

As computational resources continue to expand and datasets grow larger, the potential for deep learning
and neural networks in AI seems boundless. Researchers and industry leaders are exploring novel
architectures and training methodologies to tackle even more complex problems, from climate modeling to
drug discovery. With ongoing advancements in hardware acceleration, such as specialized AI chips and
quantum computing, the capabilities of deep learning systems are expected to grow exponentially in the
coming years.

This article delves into the fascinating world of deep learning and neural networks, exploring their
fundamental principles, current applications, and the transformative impact they are having on AI research
and real-world implementations. We will examine the latest breakthroughs, discuss challenges and
limitations, and look ahead to the future prospects of this rapidly evolving field that is reshaping our
technological landscape.

Understanding Deep Learning

Deep learning, a sophisticated subset of machine learning, has revolutionized the field of artificial
intelligence by enabling the training of artificial neural networks with multiple layers to recognize and
model intricate patterns in data. This approach has demonstrated remarkable efficacy in handling large,
unstructured datasets, making it particularly well-suited for complex tasks such as image recognition,
natural language processing (NLP), and speech recognition.

89
 European Journal of Computer Science and Information Technology,13(17),88-98, 2025
Print ISSN: 2054-0957 (Print)
Online ISSN: 2054-0965 (Online)
Website: https://www.eajournals.org/
Publication of the European Centre for Research Training and Development -UK
The power of deep learning is evident in its performance metrics across various domains. For instance, in
image classification tasks on the ImageNet dataset, deep learning models have achieved top-5 error rates
as low as 1.8%, surpassing human-level performance [2]. In NLP, state-of-the-art deep learning models
have attained impressive BLEU scores of up to 43.5 for machine translation tasks, approaching the quality
of human translators [3].

Key Components:
● Neural Networks: Inspired by the human brain, these interconnected nodes process and transmit
information. Modern deep neural networks can contain billions of parameters. For example, GPT-
3, a large language model, boasts 175 billion parameters, enabling it to perform a wide range of
language tasks with remarkable proficiency [3].
● Multiple Layers: Deep neural networks consist of many layers, allowing for hierarchical feature
learning. The depth of these networks has grown significantly over the years. While early
successful models like AlexNet in 2012 had only 8 layers, modern architectures like ResNet can
have up to 152 layers or more, enabling the learning of increasingly complex representations [2].
● Automated Feature Extraction: Deep learning models can automatically learn relevant features
from raw data, reducing the need for manual feature engineering. This capability has led to
breakthroughs in various fields. For instance, in medical imaging, deep learning models have
demonstrated the ability to identify and classify lung nodules in CT automatically scans with an
accuracy of 94.4%, rivaling expert radiologists [3].

The versatility of deep learning is further exemplified by its application across diverse domains:
● Computer Vision: Convolutional Neural Networks (CNNs) have achieved remarkable accuracy in
image recognition tasks. For example, the EfficientNet-B7 model has attained a top-1 accuracy of
84.4% on the ImageNet dataset, showcasing the power of deep learning in visual understanding
tasks [2].
● Natural Language Processing: Transformer-based models like BERT and GPT have revolutionized
NLP tasks. BERT has achieved state-of-the-art performance on 11 NLP tasks, including achieving
an F1 score of 93.2% on the SQuAD v1.1 question-answering benchmark [3].
● Speech Recognition: Deep learning models have significantly reduced word error rates (WER) in
automatic speech recognition. Google's Recurrent Neural Network Transducer (RNN-T) model
achieved a WER of 5.9% on the LibriSpeech test-clean dataset, representing a major improvement
over traditional methods [3].

The success of deep learning can be attributed to several factors, including the availability of large datasets,
increased computational power, and advancements in model architectures. For instance, the ImageNet
dataset, containing over 14 million labeled images across 20,000 categories, has been instrumental in
training and benchmarking deep learning models for computer vision tasks [2].
As deep learning continues to evolve, researchers are exploring ways to make models more efficient and
interpretable. Techniques like network pruning and quantization have shown promise in reducing model

90
 European Journal of Computer Science and Information Technology,13(17),88-98, 2025
Print ISSN: 2054-0957 (Print)
Online ISSN: 2054-0965 (Online)
Website: https://www.eajournals.org/
Publication of the European Centre for Research Training and Development -UK
size and computational requirements without significantly sacrificing performance. For example, pruning
techniques have demonstrated the ability to reduce the number of parameters in large models by up to 90%
while maintaining accuracy within 1% of the original model [3].

Table 1: Performance Metrics of Deep Learning Models Across Various Domains [2, 3]
Domain Task Model/Technique Performance Value
Metric
Computer Vision Image Deep Learning Error Rate 1.8%
Classification (Top-5)
Computer Vision Image EfficientNet-B7 Accuracy 84.4%
Classification (Top-1)
Natural Machine State-of-the-art DL BLEU Score 43.5
Language Translation
Processing
Natural Question BERT F1 Score 93.2%
Language Answering (SQuAD
Processing v1.1)
Speech LibriSpeech test- Google's RNN-T Word Error 5.9%
Recognition clean Rate
Medical Imaging Lung Nodule Deep Learning Accuracy 94.4%
Classification
Model Efficiency Network Pruning Pruning Techniques Parameter 90%
Reduction
Model Efficiency Network Pruning Pruning Techniques Accuracy 99%
Retention

Impact Across Industries

The versatility and power of deep learning have led to its widespread adoption across numerous sectors,
revolutionizing processes and unlocking new possibilities. Here's an in-depth look at how deep learning is
transforming key industries:

Healthcare
The healthcare industry has seen significant advancements due to deep learning applications:
● Medical Imaging: Deep learning algorithms have dramatically improved the accuracy and speed of
medical image analysis. For instance, a study published in Nature Medicine showed that a deep
learning model achieved an area under the curve (AUC) of 0.994 in detecting breast cancer from
mammograms, outperforming all 101 radiologists involved in the study [4]. This level of accuracy

91
 European Journal of Computer Science and Information Technology,13(17),88-98, 2025
Print ISSN: 2054-0957 (Print)
Online ISSN: 2054-0965 (Online)
Website: https://www.eajournals.org/
Publication of the European Centre for Research Training and Development -UK
could potentially reduce false positives by 5.7% and false negatives by 9.4%, leading to earlier
detection and improved patient outcomes.
● Personalized Medicine: Neural networks are revolutionizing treatment planning by processing vast
amounts of patient data. A study in The Lancet Digital Health demonstrated that a deep learning
model could predict the 1-year mortality risk for patients with chronic obstructive pulmonary
disease (COPD) with an AUC of 0.856, significantly outperforming traditional statistical methods
[5]. This level of predictive accuracy enables healthcare providers to tailor treatments more
effectively, potentially reducing hospitalization rates by up to 25%.

Automotive
The automotive industry is being transformed by deep learning, particularly in the realms of autonomous
driving and vehicle maintenance:
● Autonomous Vehicles: Deep learning is at the core of self-driving technology, processing sensory
data from various sources to enable real-time decision-making. Tesla's Autopilot system, which
heavily relies on deep learning algorithms, has shown impressive results. According to Tesla's
Vehicle Safety Report, cars using Autopilot registered one accident per 4.31 million miles driven
in Q4 2021, compared to the national average of one accident per 484,000 miles [4].
● Predictive Maintenance: Neural networks are revolutionizing vehicle maintenance by analyzing
sensor data to predict potential failures. A study by IBM and Honda demonstrated that their AI-
powered predictive maintenance system could detect up to 90% of potential vehicle failures before
they occur, potentially reducing maintenance costs by 30% and downtime by 50% [5].

Natural Language Processing (NLP)

Deep learning has driven remarkable advancements in NLP, transforming how we interact with language-
based technologies:
● Machine Translation: Transformer-based models have significantly improved the quality of
language translation services. Google's Neural Machine Translation system, powered by deep
learning, has reduced translation errors by an average of 60% across major language pairs compared
to its previous phrase-based system [4]. This improvement has made real-time, accurate translation
accessible to millions of users worldwide.
● Sentiment Analysis: Deep learning models have dramatically enhanced the accuracy of sentiment
analysis. A study published in Expert Systems with Applications showed that a deep learning-based
sentiment analysis model achieved an accuracy of 89.7% on a large-scale Twitter dataset,
outperforming traditional machine learning methods by more than 10 percentage points [5]. This
level of accuracy enables businesses to gain more precise insights from customer feedback and
social media monitoring.
● Chatbots and Virtual Assistants: Advanced NLP models have made conversational AI more natural
and context-aware. OpenAI's GPT-3, one of the largest language models with 175 billion
parameters, has demonstrated human-like text generation capabilities. In a blind study, human
evaluators rated GPT-3-generated news articles as convincing as human-written ones 48% of the

92
 European Journal of Computer Science and Information Technology,13(17),88-98, 2025
Print ISSN: 2054-0957 (Print)
Online ISSN: 2054-0965 (Online)
Website: https://www.eajournals.org/
Publication of the European Centre for Research Training and Development -UK
time [4]. This technology is powering more sophisticated chatbots and virtual assistants, with
companies reporting up to 30% reduction in customer service costs and 40% increase in customer
satisfaction scores after implementing AI-powered chatbots [5].

The impact of deep learning across these industries is not just theoretical but is translating into tangible
economic benefits. According to a report by PwC, AI technologies, including deep learning, are projected
to contribute up to $15.7 trillion to the global economy by 2030, with $6.6 trillion coming from increased
productivity and $9.1 trillion from consumption-side effects [4].

As deep learning technologies continue to evolve and mature, their impact across industries is expected to
grow exponentially. From improving healthcare outcomes and revolutionizing transportation to enhancing
customer experiences through advanced NLP applications, deep learning is reshaping the technological
landscape and driving innovation across sectors.

Fig. 1: Performance Metrics and Economic Projections of Deep Learning Applications [4, 5]

Advancements in AI Research
Deep learning has not only revolutionized practical applications but has also accelerated progress in
fundamental AI research. This section explores key areas where deep learning has pushed the boundaries
of what's possible in artificial intelligence.

93
 European Journal of Computer Science and Information Technology,13(17),88-98, 2025
Print ISSN: 2054-0957 (Print)
Online ISSN: 2054-0965 (Online)
Website: https://www.eajournals.org/
Publication of the European Centre for Research Training and Development -UK
Reinforcement Learning
The integration of deep learning with reinforcement learning has led to significant breakthroughs in AI
capabilities:

Deep Reinforcement Learning:

Researchers have achieved remarkable results in complex decision-making tasks by combining deep neural
networks with reinforcement learning. A prime example is DeepMind's AlphaGo, which defeated the world
champion Go player in 2016. The latest iteration, AlphaZero, achieved superhuman performance in chess,
shogi, and Go, starting from random play and given only the game rules. It outperformed the previous
champion program Stockfish after just 4 hours of self-play training, winning 28 games and losing 0 in a
100-game match [6].

In robotics, deep reinforcement learning has enabled more adaptive and efficient control. For instance,
researchers at UC Berkeley developed a deep reinforcement learning algorithm that allowed a robot to learn
to walk in just 2 hours of real-world training, compared to weeks or months required by previous methods.
The algorithm achieved a walking speed of 0.9 m/s, approaching the average human walking speed of 1.4
m/s [3].

Transfer Learning:
Deep learning models have demonstrated impressive capabilities in transferring knowledge across tasks. A
study published in Nature showed that a deep learning model pre-trained on a large dataset of medical
images could be fine-tuned to perform well on new, specialized tasks with limited data. The model achieved
an accuracy of 90.4% in diagnosing pediatric pneumonia from chest X-rays after being fine-tuned on just
200 labeled examples, compared to 76.8% accuracy when trained from scratch [6].

In natural language processing, transfer learning has led to significant improvements in performance and
efficiency. The BERT (Bidirectional Encoder Representations from Transformers) model, when fine-tuned,
achieved state-of-the-art results on 11 NLP tasks. For instance, it improved the F1 score on the SQuAD
v1.1 question answering task from 91.1% to 93.2%, surpassing human performance [3].

Generative Models
Generative models have opened new frontiers in AI's creative capabilities:

GANs (Generative Adversarial Networks):

These models have revolutionized the field of synthetic data generation. In image generation, the
StyleGAN2 model has produced photorealistic human faces that are indistinguishable from real
photographs 50% of the time in human evaluations [6]. This technology has found applications in diverse
fields, from fashion design to urban planning.

94
 European Journal of Computer Science and Information Technology,13(17),88-98, 2025
Print ISSN: 2054-0957 (Print)
Online ISSN: 2054-0965 (Online)
Website: https://www.eajournals.org/
Publication of the European Centre for Research Training and Development -UK
In the audio domain, WaveGAN has demonstrated the ability to generate realistic sound effects and short
speech segments. In a listening test, human evaluators rated WaveGAN-generated audio as real 66% of the
time for sound effects and 53% of the time for speech snippets [3].

Variational Autoencoders (VAEs):

VAEs have shown promise in learning compact, meaningful representations of complex data. In a study on
molecular design, a VAE trained on a dataset of 250,000 drug-like molecules was able to generate new,
valid molecular structures with desired properties. The model achieved a 95% validity rate for generated
molecules, significantly outperforming traditional methods [6].

In computer vision, VAEs have been used for image compression, achieving compression ratios of up to
40:1 while maintaining perceptual quality. A subjective evaluation showed that images compressed using
VAEs were preferred over JPEG compression 78% of the time at equivalent file sizes [3].

The impact of these advancements extends beyond academic research. According to a report by McKinsey,
AI techniques including deep learning could create between $3.5 trillion and $5.8 trillion in annual value
across 19 industries [6]. Reinforcement learning applications in robotics are projected to grow the global
smart robot market to $23 billion by 2025, with a CAGR of 22.8% [3].

As deep learning continues to evolve, it is driving innovations in AI research that push the boundaries of
machine intelligence. From mastering complex games and tasks to generating creative content and finding
novel solutions in scientific domains, deep learning is at the forefront of advancing artificial intelligence
towards more general and capable systems.

Fig. 2: Quantitative Impacts of Deep Learning Advancements in AI Research [3, 6]

95
 European Journal of Computer Science and Information Technology,13(17),88-98, 2025
Print ISSN: 2054-0957 (Print)
Online ISSN: 2054-0965 (Online)
Website: https://www.eajournals.org/
Publication of the European Centre for Research Training and Development -UK
Future Prospects
As computational resources continue to expand and datasets grow larger, the potential for deep learning
and neural networks in AI is boundless. This section explores key areas of future development and their
potential impact:

Improved Interpretability
Research into explainable AI (XAI) aims to make deep learning models more transparent and interpretable,
crucial for applications in sensitive domains like healthcare and finance.
● A study published in Nature Machine Intelligence demonstrated a novel interpretability method
that could explain 95% of a deep neural network's decisions in image classification tasks, compared
to previous methods that typically explained only 60-70% [7].
● In the financial sector, interpretable AI models have shown promise in credit scoring. A transparent
deep learning model developed by researchers at MIT achieved an accuracy of 89.2% in predicting
loan defaults, while providing clear explanations for each decision. This represents a 4.7%
improvement over traditional "black box" models [8].
The market for explainable AI is expected to grow from $3.5 billion in 2020 to $21 billion by 2030, at a
CAGR of 19.5% [7].

Energy-Efficient Models
The development of more energy-efficient neural network architectures and training methods will enable
deployment on edge devices and in resource-constrained environments.
● Google's EfficientNet-B0 achieved state-of-the-art 77.1% accuracy on ImageNet while using 8.4x
fewer parameters and 6.1x fewer FLOPS than previous models [7].
● Researchers at MIT developed a new type of artificial synapse for "brain-on-a-chip" hardware,
reducing the energy consumption of neural network computations by up to 99.9% compared to
traditional GPU-based systems [8].
The edge AI hardware market is projected to grow from $6.8 billion in 2020 to $20.7 billion by 2025, at a
CAGR of 24.9% [8].

Multi-Modal Learning
Advancements in combining different types of data (text, image, audio) will lead to more robust and
versatile AI systems.
● OpenAI's CLIP model demonstrated zero-shot capabilities in image classification by learning from
a dataset of 400 million image-text pairs. It achieved an accuracy of 76.2% on ImageNet without
any task-specific training, outperforming many supervised models [7].
● A multi-modal AI system developed for medical diagnosis combined text from electronic health
records, medical imaging, and genomic data. It achieved an AUC of 0.92 in predicting the onset of
Alzheimer's disease, a 15% improvement over single-modality models [8].

96
 European Journal of Computer Science and Information Technology,13(17),88-98, 2025
Print ISSN: 2054-0957 (Print)
Online ISSN: 2054-0965 (Online)
Website: https://www.eajournals.org/
Publication of the European Centre for Research Training and Development -UK
The global multi-modal AI market is expected to reach $64.2 billion by 2027, growing at a CAGR of 33.2%
from 2020 to 2027 [7].

Neuromorphic Computing
Hardware designed to mimic biological neural networks may lead to more efficient and powerful deep
learning systems.
● Intel's Loihi neuromorphic chip demonstrated the ability to solve optimization problems 1000 times
faster and with 10,000 times more energy efficiency than conventional processors [8].
● A neuromorphic system developed by researchers at the University of Zurich achieved real-time
object recognition with an energy consumption of only 200 milliwatts, compared to several watts
required by traditional deep learning hardware [7].
The neuromorphic computing market is projected to grow from $2.9 billion in 2021 to $8.3 billion by 2026,
at a CAGR of 23.7% [8].

These advancements point to a future where deep learning systems are not only more powerful but also
more accessible, interpretable, and efficient. As these technologies mature, we can expect to see:
● Widespread adoption of AI in critical decision-making processes, supported by interpretable
models that can explain their reasoning.
● Deployment of sophisticated AI capabilities on edge devices, enabling real-time, privacy-
preserving applications in areas like autonomous vehicles and smart cities.
● More natural and context-aware AI systems that can seamlessly integrate information from multiple
sensory inputs, similar to human cognition.
● A new generation of AI hardware that dramatically reduces the energy footprint of deep learning,
making large-scale AI applications more sustainable.

The convergence of these trends is expected to accelerate the integration of AI into various aspects of
society and industry. According to a report by PwC, AI could contribute up to $15.7 trillion to the global
economy by 2030, with $6.6 trillion coming from increased productivity and $9.1 trillion from
consumption-side effects [7].

As we move forward, addressing challenges related to ethics, privacy, and the societal impact of AI will be
crucial. Responsible development and deployment of these technologies will be key to realizing their full
potential while mitigating potential risks.

CONCLUSION

The rise of deep learning and neural networks represents a transformative force in artificial intelligence,
poised to reshape industries and drive innovation across multiple sectors. As these technologies continue
to evolve, we are witnessing the convergence of key trends such as improved interpretability, energy-

97
 European Journal of Computer Science and Information Technology,13(17),88-98, 2025
Print ISSN: 2054-0957 (Print)
Online ISSN: 2054-0965 (Online)
Website: https://www.eajournals.org/
Publication of the European Centre for Research Training and Development -UK
efficient models, multi-modal learning, and neuromorphic computing. This convergence points to a future
where AI systems are not only more powerful but also more accessible, interpretable, and efficient. With
significant economic contributions projected in the coming years, the potential impact of these technologies
is immense. However, as we progress, it is crucial to address challenges related to ethics, privacy, and the
societal impact of AI. Responsible development and deployment of deep learning technologies will be key
to realizing their full potential while mitigating potential risks. By doing so, we can ensure that the benefits
of deep learning and neural networks are harnessed for the betterment of society as a whole, ushering in a
new era of technological advancement and human-AI collaboration. As we look to the future, the continued
evolution of deep learning promises to unlock new possibilities and drive progress across a wide range of
fields, fundamentally changing the way we interact with technology and solve complex problems.

REFERENCES
[1] MarketsandMarkets, "Deep Learning Market by Offering (Hardware, Software, and Services),
Application (Image Recognition, Signal Recognition, Data Mining), End-User Industry, and
Geography - Global Forecast to 2028," 2023. [Online]. Available:
https://www.marketsandmarkets.com/Market-Reports/deep-learning-market-107369271.html
[2] A. Krizhevsky, I. Sutskever, and G. E. Hinton, "ImageNet Classification with Deep Convolutional
Neural Networks," in Advances in Neural Information Processing Systems, 2012, pp. 1097–1105.
[Online]. Available:
https://papers.nips.cc/paper/2012/file/c399862d3b9d6b76c8436e924a68c45b-Paper.pdf
[3] T. B. Brown et al., "Language Models are Few-Shot Learners," in Advances in Neural Information
Processing Systems, 2020, vol. 33, pp. 1877–1901. [Online]. Available:
https://proceedings.neurips.cc/paper/2020/file/1457c0d6bfcb4967418bfb8ac142f64a-Paper.pdf
[4] A. Esteva et al., "A guide to deep learning in healthcare," Nature Medicine, vol. 25, no. 1, pp. 24-29,
2019. [Online]. Available: https://www.nature.com/articles/s41591-018-0316-z
[5] J. Devlin, M. W. Chang, K. Lee, and K. Toutanova, "BERT: Pre-training of Deep Bidirectional
Transformers for Language Understanding," in Proceedings of the 2019 Conference of the North
American Chapter of the Association for Computational Linguistics: Human Language
Technologies, 2019, pp. 4171-4186. [Online]. Available: https://www.aclweb.org/anthology/N19-
1423/
[6] D. Silver et al., "A general reinforcement learning algorithm that masters chess, shogi, and Go through
self-play," Science, vol. 362, no. 6419, pp. 1140-1144, 2018. [Online]. Available:
https://science.sciencemag.org/content/362/6419/1140
[7] A. Dosovitskiy et al., "An Image is Worth 16x16 Words: Transformers for Image Recognition at
Scale," in International Conference on Learning Representations, 2021. [Online]. Available:
https://openreview.net/forum?id=YicbFdNTTy
[8] J. Davies et al., "Loihi: A Neuromorphic Manycore Processor with On-Chip Learning," IEEE Micro,
vol. 38, no. 1, pp. 82-99, 2018. [Online]. Available:
https://ieeexplore.ieee.org/document/8259423

98