2023 10th IEEE Uttar Pradesh Section International Conference on Electrical, Electronics and Computer Engineering (UPCON)

Neuro-symbolic AI: Integrating Symbolic

2023 10th IEEE Uttar Pradesh Section International Conference on Electrical, Electronics and Computer Engineering (UPCON) | 979-8-3503-8247-1/23/$31.00 ©2023 IEEE | DOI: 10.1109/UPCON59197.2023.10434380

Reasoning with Deep Learning

Modi Himabindu Revathi V Manish Gupta
Institute of Aeronautical Engineering, Department of Applied Sciences, Lovely Professional University,
Dundigal, Hyderabad, India. New Horizon College of Engineering, Phagwara, India;
[email protected] Bangalore, [email protected]
[email protected]

Dr. Ajay Rana Pradeep Kumar Chandra Hayder Saadoon Abdulaali

Director General Lloyd Institute of Engineering & College of Engineering Technology,
Amity University, Technology, National University of Science and
Greater Noida Knowledge Park II, Greater Noida, Uttar Technology,
[email protected] Pradesh, Dhi Qar, Iraq;
[email protected] [email protected]

Abstract-- Neuro-symbolic artificial intelligence (AI) stands symbolic reasoning — a paradigm of AI that relies on logic
at the frontier of machine learning by amalgamating the and well-defined symbols to perform reasoning tasks,
interpretability and structured knowledge representation of providing transparency and interpretability at the cost of
symbolic reasoning with the adaptive learning capabilities of flexibility and scalability [3]. The central thesis of this paper
deep neural networks. This paper presents a comprehensive posits that a synergistic amalgamation of symbolic reasoning
framework for neuro-symbolic integration, outlining a and deep learning, herein referred to as neuro-symbolic AI,
harmonized architecture that leverages the strengths of both can yield a class of models that inherit the strengths of both
domains. The proposed system utilizes symbolic AI to impose approaches while mitigating their individual weaknesses.
structural constraints and inject domain knowledge into the
This integration aspires to construct AI systems capable of
learning process, enhancing the reasoning capabilities of deep
learning models. Concurrently, it capitalizes on the proficiency
robust reasoning, generalization, and knowledge transfer,
of deep learning in handling high-dimensional, noisy data, operating with a level of cognitive adeptness reminiscent of
enabling the symbolic components to operate beyond discrete, human intelligence [4]. The potential of neuro-symbolic AI
well-defined environments. The architecture is validated lies in its ability to imbue deep learning systems with the
through a series of experiments demonstrating enhanced capacity for symbolic manipulation and structured
performance in tasks requiring complex reasoning, knowledge representation, facilitating a more profound
generalization, and knowledge transfer. The framework comprehension of the tasks at hand. Symbolic AI, with its
showcases a significant reduction in data dependency for roots stretching back to the inception of the field, provides a
model training, increased interpretability of the decision- framework for knowledge representation that is explicit,
making process, and robustness to noise and ambiguity. This interpretable, and amenable to manipulation based on logical
integration marks a stride towards the development of AI rules [5]. It has excelled in areas where the domain
systems with advanced cognitive abilities, akin to human-like knowledge is well-understood and can be codified into clear,
understanding and reasoning. The paper concludes with a deterministic rules. However, symbolic systems are
discussion on the implications of neuro-symbolic AI in notoriously brittle; they struggle with the ambiguity and
advancing the field and its potential to transform future AI variability inherent in real-world data, and their reliance on
applications. hand-crafted features and rules poses limitations on their
scalability and adaptability [6]. Conversely, deep learning, a
Keywords— Neuro-symbolic AI, Symbolic Reasoning, Deep
subset of machine learning characterized by multi-layered
Learning, Knowledge Representation, Cognitive AI Systems.
neural networks, has proven adept at digesting large datasets,
I. INTRODUCTION uncovering intricate patterns, and learning representations in
an end-to-end manner. These models have set benchmarks
The emergence of AI has engendered transformative across various tasks, from vision and language processing to
changes across a multitude of domains, propelling complex games like Go and Chess. Yet, their success comes
advancements that seemed infeasible merely a decade ago with caveats: they are often data-hungry, their learned
[1]. This progression is rooted in the prolific development of representations are not explicitly understandable, and they
machine learning algorithms, particularly deep learning lack the ability to reason abstractly or to transfer learned
models, which have demonstrated exceptional capabilities in concepts readily between disparate tasks [7]. Neuro-
learning representations and patterns from vast amounts of symbolic AI seeks to bridge these gaps by constructing
data. However, despite their prowess, these models often models that combine the data-driven, pattern-recognition
remain inscrutable black boxes, offering little in the way of abilities of neural networks with the explicit, rule-based
interpretability or understanding of the underlying decision- reasoning of symbolic AI. The premise is that symbolic
making processes [2]. Furthermore, they require substantial reasoning can guide neural networks to learn more structured
volumes of data to learn effectively, struggle with and generalizable representations, while neural networks can
transferring knowledge across domains, and falter in endow symbolic systems with the ability to handle noisy,
scenarios demanding intricate logical reasoning. In response unstructured data [8]. The significance of this integration is
to these challenges, there has been a resurgent interest in

979-8-3503-8247-1/23/$31.00 ©2023 IEEE 1587

Authorized licensed use limited to: Universita degli Studi di Napoli Federico II. Downloaded on August 08,2024 at 10:49:39 UTC from IEEE Xplore. Restrictions apply.
2023 10th IEEE Uttar Pradesh Section International Conference on Electrical, Electronics and Computer Engineering (UPCON)

manifold. By combining the two paradigms, the proposed for effective training [20]. Moreover, the combination of
framework aims to reduce the data requirements for training rules and embeddings via NeSy for knowledge base
AI systems, as symbolic rules can provide a priori completion has shown that not all rule-based models are the
knowledge that would otherwise need to be learned from same, with distinct approaches learning different aspects
data.[35]. This is particularly vital in domains where data is such as relations or paths [21]. Despite these advancements,
scarce or expensive to obtain. Moreover, the interpretability gaps in current methodologies persist. One of the main
inherent in symbolic systems can be infused into deep challenges is the scalability of NeSy systems to complex
learning models, enabling stakeholders to understand and real-world problems, where the amount of data and the
trust the decisions made by AI, which is critical in sensitive complexity of relationships can be overwhelming [22]. The
applications such as healthcare and criminal justice. Lastly, balance between the expressiveness of symbolic
the ability of symbolic AI to perform logical reasoning and representations and the generalizability of neural networks is
abstraction can significantly enhance the cognitive delicate and often difficult to achieve [23]. Furthermore, the
capabilities of neural networks, allowing for more deployment of NeSy in safety-critical applications, such as
sophisticated and human-like problem-solving abilities.[36] human performance prediction, requires addressing
outstanding challenges and proposing viable solutions [24].
The neuro-symbolic AI framework proposed herein The application of NeSy in smart cities advocates for a
marks a step towards reconciling the dichotomy between complete integration of neural and symbolic AI, compatible
data-driven and rule-based AI approaches. It serves as a with standard software, pointing towards a future where AI
blueprint for developing intelligent systems that can navigate can be both intelligent and interpretable [25]. However, the
the complexities of real-world data while retaining the ability realization of this vision necessitates overcoming the current
to reason and generalize in a manner akin to human limitations, such as the difficulty in integrating probabilistic
cognition.[37]. This paper will detail the theoretical reasoning with logic and neural networks, an integration
underpinnings of the framework, the architecture and exemplified by frameworks like DeepProbLog [26].
integration mechanisms, and the empirical evaluations that
underscore its efficacy. Through this exploration, it aims to While neuro-symbolic AI holds the promise of creating
chart a course for the future of AI, where the confluence of more intelligent and interpretable AI systems, significant
learning and reasoning becomes the cornerstone of intelligent research gaps remain. Addressing these gaps requires
systems.[38]. innovative approaches to knowledge representation, model
scalability, and the integration of probabilistic reasoning. As
II. LITERATURE REVIEW the field progresses, it is imperative to continue exploring the
Neuro-symbolic AI (NeSy) represents a paradigm shift in synergies between neural and logical components to unlock
artificial intelligence, aiming to bridge the gap between the full potential of AI [27].
neural networks' learning capabilities and symbolic AI's
logical reasoning. Traditional AI approaches have been III. NEURO-SYMBOLIC INTEGRATION FRAMEWORK
predominantly symbolic, relying on logic-based systems that The proposed neuro-symbolic integration framework
are interpretable but lack the ability to learn from data. endeavors to construct a cohesive model that amalgamates
Neural networks, on the other hand, excel at learning the representational benefits of symbolic reasoning with the
complex patterns from large datasets but are often criticized learning proficiencies of deep neural networks. This
for their lack of interpretability and reasoning capabilities [9- integration is predicated on a bi-directional architecture
13]. Neuro-symbolic AI seeks to combine the strengths of where symbolic reasoning informs the structure and function
both approaches, creating systems that can learn from data of neural networks, and neural networks enhance the
while also reasoning about the learned knowledge [14]. The applicability of symbolic reasoning to unstructured data [28].
integration of symbolic reasoning into neural networks has At the core of this framework is the symbiotic layer, a novel
led to systems where logic is compiled into the neural construct designed to facilitate the seamless exchange of
architecture, satisfying more goals of NeSy, such as information between the neural and symbolic components.
interpretability and adaptability [15]. However, the challenge The layer operates by translating the activation patterns of
remains in how to represent knowledge effectively and how neural networks into symbolic expressions and vice versa
to choose the appropriate neural architecture for a given task [29]. This translation is not a mere mapping but an adaptive
[16]. The infusion of external, expert-curated knowledge into process that evolves with the learning progression of the
data-driven learning methodologies has been explored to network, ensuring that the symbolic expressions remain
enhance consistency and robustness in outcomes, particularly relevant, and the neural activations reflect enhanced
in fields like natural language processing and computer reasoning capabilities. The symbolic component is structured
vision [17]. At the industrial level, companies like Bosch around a dynamic knowledge base, encoded using a formal
have exemplified the use of NeSy, where semantic logic system that can encompass a variety of logical
technologies are crucial for unifying heterogeneous data into frameworks such as first-order or description logics.[39]. The
uniform formats, facilitating better decision-making knowledge base is not static; it is subject to refinement and
processes [18]. The performance characteristics of neuro- expansion as the system interacts with data. Logic inference
symbolic models have been scrutinized, revealing that mechanisms are embedded within this component, enabling
symbolic models exhibit less potential parallelism than the system to perform deductive, inductive, and abductive
traditional neural models due to complex control flows and reasoning.[40]. These mechanisms are instrumental in
operations with low operational intensity [19]. In the context generating hypotheses and explanations that can be tested
of mental healthcare, neuro-symbolic methods have been and validated through interaction with the environment. To
investigated for infusing clinical knowledge to improve the imbue the neural network with the ability to utilize the
outcomes of neural-AI systems, demonstrating the utility of knowledge represented symbolically, an interface is
diverse clinical knowledge in creating specialized datasets introduced that aligns the probabilistic learning of the

1588
Authorized licensed use limited to: Universita degli Studi di Napoli Federico II. Downloaded on August 08,2024 at 10:49:39 UTC from IEEE Xplore. Restrictions apply.
2023 10th IEEE Uttar Pradesh Section International Conference on Electrical, Electronics and Computer Engineering (UPCON)

network with the deterministic logic of the symbolic system

[30]. This is achieved by formulating a set of constraints
derived from the symbolic knowledge that guides the
network's learning process. These constraints are
incorporated into the loss function of the network, which is
then optimized using a gradient-based learning algorithm.
This process ensures that the network’s learning trajectory is
consistent with the domain knowledge encapsulated by the
symbolic component. The learning algorithm itself is a
hybrid, accommodating both the backpropagation of errors
for neural adjustments and the refinement of symbolic rules
based on feedback from the neural component. This bi-
directional learning paradigm enables the system to not only
learn from data but also to adapt its symbolic reasoning in
light of empirical evidence. This adaptability is crucial for
the system to remain applicable in dynamically changing
environments and for tasks that require ongoing learning
[31]. To address the challenge of interpretability, the
framework incorporates a mechanism for explicating the
decision-making process.

Fig. 2. Neuro-Symbolic Information Exchange Process

The architecture also includes a novel memory module

that stores instances where the integration of neural and
symbolic processing led to successful outcomes. This
historical data acts as a reference for the system to identify
patterns in the integration process that yield positive results,
thus informing future interactions between the neural and
symbolic components. An important aspect of the framework
is its scalability. The architecture is designed to be modular,
allowing for the incorporation of additional neural or
symbolic modules as required by the complexity of the task.
This modularity enables the system to be tailored to specific
applications, from simple classification tasks to complex
Fig. 1. Neuro-Symbolic Integration Framework problem-solving scenarios that necessitate advanced
reasoning capabilities. In operational terms, the framework is
This mechanism leverages the symbolic component to instantiated through a series of stages [33]. Initially, the
provide human-understandable explanations for the actions neural network is pre-trained on relevant data to establish a
and decisions of the neural network. By doing so, the system baseline of performance. Subsequently, the symbolic
can justify its conclusions in a manner that can be audited reasoning is integrated, refining the network’s capabilities
and scrutinized by human experts, which is a significant step and imposing structure on the learned representations. The
towards transparent AI systems [32]. Figure 1 illustrates the system then enters a phase of joint optimization, where the
architecture of the neuro-symbolic integration framework. It neural and symbolic components are co-trained to achieve
consists of three main components: the Neural Network, the harmonious functionality. Finally, the system is subjected to
Symbolic Reasoning module, and the Symbiotic Layer (See a series of evaluations on tasks that require both deep
Figure 2). learning and symbolic reasoning, ensuring that the

1589
Authorized licensed use limited to: Universita degli Studi di Napoli Federico II. Downloaded on August 08,2024 at 10:49:39 UTC from IEEE Xplore. Restrictions apply.
2023 10th IEEE Uttar Pradesh Section International Conference on Electrical, Electronics and Computer Engineering (UPCON)

integration leads to tangible improvements in performance with neural learning, affirming the hypothesis that such
[34]. To summarize, the proposed neuro-symbolic integration is beneficial for complex cognitive tasks. The
integration framework presents an innovative architecture discussion delves into the interpretation of these results,
that integrates the structured reasoning capabilities of attributing the performance gains to several key advantages
symbolic AI with the adaptive learning potential of neural of the neuro-symbolic approach.
networks. Through its bi-directional learning and reasoning
processes, modular design, and focus on interpretability, the
framework provides a comprehensive solution to the
challenges facing AI systems, paving the way for the
development of intelligent systems with enhanced cognitive
capabilities.
IV. EXPERIMENTAL VALIDATION AND RESULTS
The efficacy of the proposed neuro-symbolic integration
framework is substantiated through a comprehensive
experimental validation. The validation process encompasses
a multi-faceted simulation setup designed to evaluate the
system’s reasoning, generalization, and knowledge transfer
capabilities across varied tasks. The simulation setup is Fig. 3. Performance Comparison on Cognitive Tasks
structured to mirror complex real-world scenarios,
employing datasets with inherent ambiguity and noise. The Firstly, the structured knowledge provided by symbolic
datasets span across different domains, including visual reasoning aids the neural network in focusing its learning on
reasoning tasks on synthetic datasets like CLEVR, language relevant patterns, thus improving efficiency. Secondly, the
understanding benchmarks such as SQuAD, and tabular data ability of the symbolic component to perform logical
from UCI Machine Learning Repository for relational operations enables the system to handle tasks that require
reasoning. Each dataset presents unique challenges that test more than mere pattern recognition, such as inference and
the framework's ability to leverage symbolic reasoning deduction. Thirdly, the system's interpretability is enhanced,
within a neural learning environment. A baseline neural as decisions can be traced back to symbolic rules that are
network model, consisting of a CNN for visual tasks, a RNN understandable to human operators. Moreover, the
for sequential data, and a fully connected network for tabular generalization capabilities of the model are scrutinized by
data, is established for comparison.The results are presented testing its performance on data distributions that differ from
in a tabular format, delineating the performance metrics the training set. The neuro-symbolic model's ability to
across tasks for both baseline and neuro-symbolic models. maintain high performance in these tests suggests that the
Performance metrics include accuracy, F1 score, and mean symbolic rules provide a form of inductive bias that guides
reciprocal rank, providing a holistic view of the system's the network towards learning more generalizable features.
capabilities. For visual reasoning tasks, the neuro-symbolic The validation process also examines the robustness of the
model demonstrates a marked improvement in accuracy over model to noise and perturbations in the data. The neuro-
the baseline, especially in scenarios requiring compositional symbolic model demonstrates resilience, attributed to the
reasoning and generalization to unseen combinations of symbolic component's capacity to enforce logical
objects and attributes. In language understanding, the neuro- consistency, thereby providing a counterbalance to the
symbolic model shows superior performance in question network's susceptibility to overfitting on noisy data. In
answering accuracy, owing to its ability to incorporate addition to performance metrics, the simulation setup
structured knowledge into the learning process. In relational includes qualitative evaluations, where the model's reasoning
reasoning tasks, the model exhibits enhanced F1 scores, processes are manually inspected to assess the
indicating its proficiency in discerning and applying interpretability of its decisions. The symbolic explanations
relational rules. The results are tabled In Table 1 and generated by the model align with human reasoning patterns,
illustrated in Figure 3. indicating that the model's decisions are not only accurate but
also grounded in logical principles. The observed
TABLE I. COMPARATIVE PERFORMANCE OF BASELINE AND NEURO- performance gains affirm the potential of neuro-symbolic
SYMBOLIC MODELS ON COGNITIVE TASKS integration to produce AI systems with enhanced reasoning
Task Metric Baseline Model Neuro- and learning abilities.
Symbolic
Model V. CONCLUSION
Visual Accuracy 75.2% 92.5%
Reasoning The exploration of neuro-symbolic integration presented
(CLEVR) in this paper offers compelling evidence for its potential to
Language F1 Score 80.1 88.7 revolutionize the field of artificial intelligence. The
Understanding integration of symbolic reasoning with deep learning not
(SQuAD) only enhances performance metrics but also imbues AI
Relational Mean 0.679 0.823 systems with a more profound level of interpretability and
Reasoning Reciprocal
(UCI) Rank robustness. The results from the visual reasoning, language
understanding, and relational reasoning tasks collectively
demonstrate that the neuro-symbolic models excel in
These results showcase the significant uplift in scenarios requiring complex and abstract reasoning, a
performance metrics when symbolic reasoning is integrated domain where conventional deep learning models often
falter. The performance gains are not merely incremental;

1590
Authorized licensed use limited to: Universita degli Studi di Napoli Federico II. Downloaded on August 08,2024 at 10:49:39 UTC from IEEE Xplore. Restrictions apply.
2023 10th IEEE Uttar Pradesh Section International Conference on Electrical, Electronics and Computer Engineering (UPCON)

they represent significant strides in the model's ability to [9] Radhakrishna, V., Kumar, P.V., Janaki, V., Rajasekhar, N. (2017).
generalize and reason in a manner that parallels human Estimating Prevalence Bounds of Temporal Association Patterns to
Discover Temporally Similar Patterns. In: Matoušek, R. (eds) Recent
cognitive processes. Furthermore, the robustness of the Advances in Soft Computing. ICSC-MENDEL 2016. Advances in
neuro-symbolic model to noisy and perturbed data suggests a Intelligent Systems and Computing, vol 576. Springer, Cham.
resilience that is critical for real-world applications. AI https://doi.org/10.1007/978-3-319-58088-3_20
systems deployed in dynamic environments must be capable [10] Kalyani G., Janakiramaiah B., Karuna A., Prasad L.V.N., Diabetic
of maintaining performance despite the inevitable variability retinopathy detection and classification using capsule networks,
and imperfections in the data they encounter. The neuro- Complex and Intelligent Systems, 2023, 10.1007/s40747-021-00318-
9
symbolic framework addresses this need, suggesting a
[11] A. Cheruvu, V. Radhakrishna and N. Rajasekhar, "Using normal
pathway to more reliable and trustworthy AI. distribution to retrieve temporal associations by Euclidean distance,"
The added interpretability that comes with symbolic 2017 International Conference on Engineering & MIS (ICEMIS),
reasoning cannot be overstated. As AI continues to permeate Monastir, Tunisia, 2017, pp. 1-3, doi:
sensitive sectors, the demand for transparent and explainable 10.1109/ICEMIS.2017.8273101.
systems will only escalate. The ability of neuro-symbolic [12] Awasthi, Ankita, and Kuldeep K. Saxena. "Evaluation of mechanical
models to provide human-understandable rationales for their properties of orange peel reinforced epoxy composite." Materials
Today: Proceedings 18 (2019): 3821-3826.
decisions meets this demand, enabling stakeholders to
[13] Bisht, Pankaj Singh, and Ankita Awasthi. "Design and Analysis of
validate and trust AI outputs. Composite and Al Alloy Wheel Rim." In Advances in Materials
The potential of neuro-symbolic AI to adapt and learn in Engineering and Manufacturing Processes: Select Proceedings of
continually evolving environments presents an exciting ICFTMM 2019, pp. 15-29. Springer Singapore, 2020.
frontier for further exploration. Additionally, the scalability [14] Awasthi, Ankita, Kuldeep K. Saxena, and Vanya Arun.
of the proposed framework invites investigation into its "Sustainability and survivability in manufacturing sector." In Modern
application across an even broader array of tasks, particularly Manufacturing Processes, pp. 205-219. Woodhead Publishing, 2020.
those that have resisted previous AI approaches. The neuro- [15] Bisht, Pankaj Singh, and Ankita Awasthi. "Analysis of E-glass fiber
symbolic integration framework heralds a significant wheel rim by using ANSYS." In Recent Advances in Mechanical
Engineering: Select Proceedings of ITME 2019, pp. 79-91. Springer
advancement towards the creation of AI systems that can Singapore, 2021.
learn, reason, and explain in ways that were previously [16] Awasthi, Ankita, Kuldeep K. Saxena, and Ravi K. Dwivedi. "An
unattainable. The findings presented advocate for a investigation on classification and characterization of bio materials
continued pursuit of this integration, with the ultimate goal and additive manufacturing techniques for bioimplants." Materials
of developing AI that can seamlessly collaborate with Today: Proceedings 44 (2021): 2061-2068.
humans to solve the complex problems of the future. [17] Awasthi, Ankita, Kuldeep K. Saxena, and Vanya Arun. "Sustainable
and smart metal forming manufacturing process." Materials Today:
REFERENCES Proceedings 44 (2021): 2069-2079.
[18] Awasthi, Ankita, Akash Gupta, Kuldeep K. Saxena, and Ravi K.
[1] N. Jaidass, C. Krishna Moorthi, A. Mohan Babu, M. Reddi Babu,
Diwedi. "Equal channel angular processing on aluminium and its
Luminescence properties of Dy3+ doped lithium zinc borosilicate
alloys–A review." Materials Today: Proceedings 56 (2022): 2388-
glasses for photonic applications, Heliyon, Volume 4, Issue 3, 2018,
2391.
e00555, ISSN 2405-8440,
https://doi.org/10.1016/j.heliyon.2018.e00555. [19] Awasthi, Ankita, U. Sathish Rao, Kuldeep K. Saxena, and Ravi K.
Dwivedi. "Impact of equal channel angular pressing on aluminium
[2] Spandana K., Rao V.R.S., Internet of Things (Iot) Based smart water
alloys: An overview." Materials Today: Proceedings 57 (2022): 908-
quality monitoring system, International Journal of Engineering and
912.
Technology (UAE), 2018, 7, 3, 259-262, 10.14419/ijet.v7i3.6.14985
[20] Awasthi, Ankita, Kuldeep K. Saxena, R. K. Dwivedi, Dharam
[3] Ch. Usha Kumari, A. Sampath Dakshina Murthy, B. Lakshmi
Buddhi, and Kahtan A. Mohammed. "Design and analysis of ECAP
Prasanna, M. Pala Prasad Reddy, Asisa Kumar Panigrahy, An
Processing for Al6061 Alloy: a microstructure and mechanical
automated detection of heart arrhythmias using machine learning
property study." International Journal on Interactive Design and
technique: SVM, Materials Today: Proceedings, Volume 45, Part 2,
Manufacturing (IJIDeM) (2022): 1-13.
2021, Pages 1393-1398, ISSN 2214-7853,
https://doi.org/10.1016/j.matpr.2020.07.088. [21] Awasthi, Ankita, Akash Gupta, Kuldeep K. Saxena, R. K. Dwivedi,
Deepak Kundalkar, Dalael Saad Abdul-Zahra, Abhishek Joshi, and H.
[4] J Suresh Goud, Pudhari Srilatha, R.S. Varun Kumar, K. Thanesh
S. Saggu. "Design and analysis of equal-channel angular pressing of
Kumar, Umair Khan, Zehba Raizah, Harjot Singh Gill, Ahmed M.
Al6061: a comparative study." Advances in Materials and Processing
Galal, Role of ternary hybrid nanofluid in the thermal distribution of a
Technologies (2022): 1-10.
dovetail fin with the internal generation of heat, Case Studies in
Thermal Engineering,Volume 35, 2022, 102113, ISSN 2214-157X, [22] Tripathi, Gyan Prakash, Sumit Agarwal, Ankita Awasthi, and Vanya
https://doi.org/10.1016/j.csite.2022.102113. Arun. "Artificial Hip Prostheses Design and Its Evaluation by Using
Ansys Under Static Loading Condition." In Biennial International
[5] Basavapoornima C., Kesavulu C.R., Maheswari T., Pecharapa W.,
Conference on Future Learning Aspects of Mechanical Engineering,
Depuru S.R., Jayasankar C.K., Spectral characteristics of Pr3+-doped
pp. 815-828. Singapore: Springer Nature Singapore, 2022.
lead based phosphate glasses for optical display device applications,
Journal of Luminescence, 2020, 228, 10.1016/j.jlumin.2020.117585 [23] Arun, V., N. K. Shukla, A. K. Singh, and K. K. Upadhyay. "Design of
all optical line selector based on SOA for Data Communication:
[6] Ramu, G. A secure cloud framework to share EHRs using modified
Proceedings of the Sixth International Conference on Computer and
CP-ABE and the attribute bloom filter. Educ Inf Technol 23, 2213–
Communication Technology 2015." In ACM Other conferences.
2233 (2018). https://doi.org/10.1007/s10639-018-9713-7
2015.
[7] Nagarjuna T., Nehru K., Nagendra Prasad G., Menakadevi N., Smart
[24] Arun, Vanya, Ashutosh Kr Singh, N. K. Shukla, and D. K. Tripathi.
sensor network based high quality air pollution monitoring system
"Design and performance analysis of SOA–MZI based reversible
using labview, International Journal of Online Engineering, 2017, 13,
toffoli and irreversible AND logic gates in a single photonic
8, 79-87, 10.3991/ijoe.v13i08.7161
circuit." Optical and quantum electronics 48 (2016): 1-15.
[8] Indira DNVSLS, Ganiya RK, Ashok Babu P, Xavier AJ, Kavisankar
[25] Arun, Vanya, Kapil Deo Bodha, Awadhesh K. Maurya, and Ashutosh
L, Hemalatha S, Senthilkumar V, Kavitha T, Rajaram A, Annam K,
K. Singh. "Design and implementation of all optical processing units
Yeshitla A. Improved Artificial Neural Network with State Order
together performing arithmetic and logical functions." In VLSI,
Dataset Estimation for Brain Cancer Cell Diagnosis. Biomed Res Int.
Microwave and Wireless Technologies: Select Proceedings of
2022 Apr 16;2022:7799812. doi: 10.1155/2022/7799812. PMID:
ICVMWT 2021, pp. 83-93. Singapore: Springer Nature Singapore,
35480141; PMCID: PMC9038414.
2022.

1591
Authorized licensed use limited to: Universita degli Studi di Napoli Federico II. Downloaded on August 08,2024 at 10:49:39 UTC from IEEE Xplore. Restrictions apply.
2023 10th IEEE Uttar Pradesh Section International Conference on Electrical, Electronics and Computer Engineering (UPCON)

[26] Swapna Sri, M. Naga, P. Anusha, VV venu Madhav, Kuldeep Kumar

Saxena, Ch Sri Chaitanya, R. Haranath, and Bharat Singh. "Influence [35] S. Sharma, M. M. Tripathi and V. M. Mishra, "Comparative Analysis
of Cu particulates on a356mmc using frequency response function of Routing Protocols in Wireless Body Area Network (WBAN),"
and damping ratio." Advances in Materials and Processing 2022 2nd International Conference on Innovative Practices in
Technologies (2023): 1-9. Technology and Management (ICIPTM), Gautam Buddha Nagar,
[27] Arora, Gurmeet Singh, and Kuldeep Kumar Saxena. "A review study India, 2022, pp. 703-706, doi: 10.1109/ICIPTM54933.2022.9754202.
on the influence of hybridization on mechanical behaviour of hybrid [36] N. Krishnachaithanya et al., "People Counting in Public Spaces using
Mg matrix composites through powder metallurgy." Materials Today: Deep Learning-based Object Detection and Tracking Techniques,"
Proceedings (2023). 2023 International Conference on Computational Intelligence and
[28] Bodha, Kapil Deo, V. Mukherjee, and Vinod Kumar Yadav. "A Sustainable Engineering Solutions (CISES), Greater Noida, India,
player unknown's battlegrounds ranking based optimization technique 2023, pp. 784-788, doi: 10.1109/CISES58720.2023.10183503.
for power system optimization problem." Evolving Systems 14, no. 2 [37] S. A. Yadav, S. Sharma and S. R. Kumar, "A robust approach for
(2023): 295-317. offline English character recognition," 2015 International Conference
[29] Hitzler, Pascal, Aaron Eberhart, Monireh Ebrahimi, Md on Futuristic Trends on Computational Analysis and Knowledge
Kamruzzaman Sarker, and Lu Zhou. "Neuro-symbolic approaches in Management (ABLAZE), Greater Noida, India, 2015, pp. 121-126,
artificial intelligence." National Science Review 9, no. 6 (2022): doi: 10.1109/ABLAZE.2015.7154980.
nwac035. [38] Yadav, SA, Poongodi, T. A novel chain-based clustering for green
[30] +DPLOWRQ .\OH $SDUQD 1D\DN %RMDQ %RåLü DQG /XFD /RQJR ,V communication in wireless sensor network. Int J Commun Syst. 2023;
neuro-symbolic AI meeting its promises in natural language 36(13):e5523. doi:10.1002/dac.5523
processing? A structured review." Semantic Web Preprint (2022): 1- [39] Tushar, R. K. Patel, E. Aggarwal, K. Solanki, O. Dahiya and S. A.
42. Yadav, "A Logistic Regression and Decision Tree Based Hybrid
[31] Ebrahimi, Monireh, Aaron Eberhart, Federico Bianchi, and Pascal Approach to Predict Alzheimer's Disease," 2023 International
Hitzler. "Towards bridging the neuro-symbolic gap: Deep deductive Conference on Computational Intelligence and Sustainable
reasoners." Applied Intelligence 51 (2021): 6326-6348. Engineering Solutions (CISES), Greater Noida, India, 2023, pp. 722-
[32] +DPLOWRQ .\OH $SDUQD 1D\DN %RMDQ %RåLF DQG /XFD /RQJR ,V 726, doi: 10.1109/CISES58720.2023.10183466.
Neuro-Symbolic AI Meeting its Promise in Natural Language [40] A. Chaturvedi, T. V. Kumar, A. P. Srivastava, S. K. Shukla, S. V.
Processing? A Structured." arXiv preprint arXiv:2202.12205 (2022). Singh and A. Parvez, "Energy Optimization in Wireless Sensor
[33] Shindo, Hikaru, Devendra Singh Dhami, and Kristian Kersting. Networks," 2022 2nd International Conference on Technological
"Neuro-symbolic forward reasoning." arXiv preprint Advancements in Computational Sciences (ICTACS), Tashkent,
arXiv:2110.09383 (2021). Uzbekistan, 2022, pp. 700-704, doi:
10.1109/ICTACS56270.2022.9987887.
[34] Gaur, Manas, Kalpa Gunaratna, Shreyansh Bhatt, and Amit Sheth.
"Knowledge-infused learning: A sweet spot in neuro-symbolic ai."
IEEE Internet Computing 26, no. 4 (2022): 5-11.

1592
Authorized licensed use limited to: Universita degli Studi di Napoli Federico II. Downloaded on August 08,2024 at 10:49:39 UTC from IEEE Xplore. Restrictions apply.