What Is Artificial General Intelligence?

Artificial General Intelligence (AGI) is defined as the software representation of generalized human cognitive capacities that enables the AGI system to solve problems when presented with new tasks. 

In other words, it's AI's capacity to learn similarly to humans.

Strong AI, full AI, and general intelligent action are some names for it. 

The phrase "strong AI," however, is only used in few academic publications to refer to computer systems that are sentient or aware. 

These definitions may change since specialists from many disciplines see human intelligence from various angles. 

For instance, computer scientists often characterize human intelligence as the capacity to accomplish objectives. 

On the other hand, general intelligence is defined by psychologists in terms of survival or adaptation.

Weak or narrow AI, in contrast to strong AI, is made up of programs created to address a single issue and lacks awareness since it is not meant to have broad cognitive capacities. 

Autonomous cars and IBM's Watson supercomputer are two examples. 

Nevertheless, AGI is defined in computer science as an intelligent system having full or comprehensive knowledge as well as cognitive computing skills.

As of right now, there are no real AGI systems; they are still the stuff of science fiction. 

The long-term objective of these systems is to perform as well as humans do. 

However, due to AGI's superior capacity to acquire and analyze massive amounts of data at a far faster rate than the human mind, it may be possible for AGI to be more intelligent than humans.

Artificial intelligence (AI) is now capable of carrying out a wide range of functions, including providing tailored suggestions based on prior web searches. 

Additionally, it can recognize various items for autonomous cars to avoid, recognize malignant cells during medical inspections, and serve as the brain of home automation. 

Additionally, it may be utilized to find possibly habitable planets, act as intelligent assistants, be in charge of security, and more.

Naturally, AGI seems to far beyond such capacities, and some scientists are concerned this may result in a dystopian future. 

Elon Musk said that sentient AI would be more hazardous than nuclear war, while Stephen Hawking advised against its creation because it would see humanity as a possible threat and act accordingly.

Despite concerns, most scientists agree that genuine AGI is decades or perhaps centuries away from being developed and must first meet a number of requirements (which are always changing) in order to be achieved. 

These include the capacity for logic, tact, puzzle-solving, and making decisions in the face of ambiguity. 

Additionally, it must be able to plan, learn, and communicate in natural language, as well as represent information, including common sense. 

AGI must also have the capacity to detect (hear, see, etc.) and output the ability to act, such as moving items and switching places to explore. 

How far along are we in the process of developing artificial general intelligence, and who is involved?

In accordance with a 2020 study from the Global Catastrophic Risk Institute (GCRI), academic institutions, businesses, and different governmental agencies are presently working on 72 recognized AGI R&D projects. 

According to the poll, projects nowadays are often smaller, more geographically diversified, less open-source, more focused on humanitarian aims than academic ones, and more centered in private firms than projects in 2017. 

The comparison also reveals a decline in projects with academic affiliations, an increase in projects sponsored by corporations, a rise in projects with a humanitarian emphasis, a decline in programs with ties to the military, and a decline in US-based initiatives.

In AGI R&D, particularly military initiatives that are solely focused on fundamental research, governments and organizations have very little roles to play. 

However, recent programs seem to be more varied and are classified using three criteria, including business projects that are engaged in AGI safety and have humanistic end objectives. 

Additionally, it covers tiny private enterprises with a variety of objectives including academic programs that do not concern themselves with AGI safety but rather the progress of knowledge.

One of the most well-known organizations working on AGI is Carnegie Mellon University, which has a project called ACT-R that aims to create a generic cognitive architecture based on the basic cognitive and perceptual functions that support the human mind. 

The project may be thought of as a method of describing how the brain is structured such that different processing modules can result in cognition.

Another pioneering organization testing the limits of AGI is Microsoft Research AI, which has carried out a number of research initiatives, including developing a data set to counter prejudice for machine-learning models. 

The business is also investigating ways to advance moral AI, create a responsible AI standard, and create AI strategies and evaluations to create a framework that emphasizes the advancement of mankind.

The person behind the well-known video game franchises Commander Keen and Doom has launched yet another intriguing endeavor. 

Keen Technologies, John Carmack's most recent business, is an AGI development company that has already raised $20 million in funding from former GitHub CEO Nat Friedman and Cue founder Daniel Gross. 

Carmack is one of the AGI optimists who believes that it would ultimately help mankind and result in the development of an AI mind that acts like a human, which might be used as a universal remote worker.

So what does AGI's future hold? 

The majority of specialists are doubtful that AGI will ever be developed, and others believe that the urge to even develop artificial intelligence comparable to humans will eventually go away. 

Others are working to develop it so that everyone will benefit.

Nevertheless, the creation of AGI is still in the planning stages, and in the next decades, little progress is anticipated. 

Nevertheless, throughout history, scientists have debated whether developing technologies with the potential to change people's lives will benefit society as a whole or endanger it. 

This proposal was considered before to the invention of the vehicle, during the development of AC electricity, and when the atomic bomb was still only a theory.

~ Jai Krishna Ponnappan

Find Jai on Twitter | LinkedIn | Instagram

You may also want to read more about Artificial Intelligence here.

Be sure to refer to the complete & active AI Terms Glossary here.

AI Glossary - What Is The ART 1 Algorithm?

 

What Is ART 1?

The initial Adaptive Resonance Theory (ART) model was designated as ART 1. 

It has the ability to cluster binary input variables.

What Is The Architecture And Design Of ART 1?

The Design of ART1

The following two units make up ART1:

Parameters Used Above:

n − Number of components in the input vector

m − Maximum number of clusters that can be formed

bij − Weight from F1b to F2 layer, i.e. bottom-up weights

tji − Weight from F2 to F1b layer, i.e. top-down weights

ρ − Vigilance parameter

||x|| − Norm of vector x

1. Computational Unit Of ART 1

It consists of the following:

(i) Unit of input (F1 layer) 

It also includes the next two parts:

1. F1 a layer Input portion – In ART1, this part would merely include the input vectors with no processing. It has an F1b layer interface portion connection.

2. F1b layer interface portion - The signal from the input section and the signal from the F2 layer are combined at the F1b layer interface portion. Bottom up weights bij connect the F1b layer to the F2 layer, while top down weights tji link the F2 layer to the F1b layer.

2. Cluster Unit (F2 layer): 

This is a layer that is in competition. To learn the input pattern, the unit with the highest net input is chosen. All other cluster units have their activation set to 0.

3. Reset Method: 

This mechanism operates by comparing the input vector's similarity to the top-down weight. The cluster will not be permitted to learn the pattern if the degree of similarity is less than the vigilance parameter, and a rest will take place.

4. Supplement Unit: 

In reality, the problem with the reset mechanism is that the layer F2 has to be suppressed under certain circumstances and also needs to be accessible while learning occurs. Because of this, the supplementary units G1 and G2 as well as the reset unit R were introduced. Gain control units are what they are known as. These units communicate with the other units in the network by receiving and sending signals. An inhibitory signal is denoted by a "," whereas an excitatory signal is denoted by a "+."

What Is The Adaptive Resonance Theory ART 1 Algorithm?

Step 1 − Initialize the learning rate, the vigilance parameter, and the weights as follows −

α>1and0<ρ≤1

0<bij(0)<αα−1+nandtij(0)=1

Step 2 − Continue step 3-9, when the stopping condition is not true.

Step 3 − Continue step 4-6 for every training input.

Step 4 − Set activations of all F1a and F1 units as follows

F2 = 0 and F1a = input vectors

Step 5 − Input signal from F1a to F1b layer must be sent like

si=xi

Step 6 − For every inhibited F2 node

yj=∑ibijxi the condition is yj ≠ -1

Step 7 − Perform step 8-10, when the reset is true.

Step 8 − Find J for yJ ≥ yj for all nodes j

Step 9 − Again calculate the activation on F1b as follows

xi=sitJi

Step 10 − Now, after calculating the norm of vector x and vector s, we need to check the reset condition as follows −

If ||x||/ ||s|| < vigilance parameter ρ,⁡then⁡inhibit ⁡node J and go to step 7

Else If ||x||/ ||s|| ≥ vigilance parameter ρ, then proceed further.

Step 11 − Weight updating for node J can be done as follows −

bij(new)=αxiα−1+||x||

tij(new)=xi

Step 12 − The stopping condition for algorithm must be checked and it may be as follows −

Do not have any change in weight.

Reset is not performed for units.

Maximum number of epochs reached.

 

~ Jai Krishna Ponnappan

Find Jai on Twitter | LinkedIn | Instagram

Be sure to refer to the complete & active AI Terms Glossary here.

You may also want to read more about Artificial Intelligence here.

Frequently Asked Questions:

What distinguishes ARTs 1 and 2 from one another?

The ART1 architecture is the most basic and straightforward. 

It can cluster input values with binary data. 

ART2 is an enhancement of ART1 that can cluster input data with continuous values.

What is the Process of Adaptive Resonance Theory?

A cognitive and neurological theory called adaptive resonance theory, or ART, explains how the brain develops its own ability to attend to, classify, identify, and anticipate items and events in a dynamic environment. 

The current most comprehensive set of cognitive and neurological theories for explanation and prediction is ART.

What Is The ART Network?

The ART network is essentially a vector classifier that receives an input vector and categorizes it into one of the categories based on which stored pattern it most closely matches.

What Is Fuzzy ART?

Fuzzy ART uses fuzzy set theory calculations to train the ART 1 neural network to classify solely binary input patterns.

Reference And Further Reading

• Tayyebi, S. and Soltanali, S., Fuzzy Modeling System Based on Ga Fuzzy Rule Extraction and Hybrid of Differential Evolution and Tabu Search Approaches: Application in Synthesis Gas Conversion to Valuable Hydrocarbons Process. Saeed, Fuzzy Modeling System Based on Ga Fuzzy Rule Extraction and Hybrid of Differential Evolution and Tabu Search Approaches: Application in Synthesis Gas Conversion to Valuable Hydrocarbons Process.

• Tang, Y., Qiu, J. and Gao, M., 2022. Fuzzy Medical Computer Vision Image Restoration and Visual Application. Computational and Mathematical Methods in Medicine, 2022.

• Dymora, P., Mazurek, M. and Bomba, S., 2022. A Comparative Analysis of Selected Predictive Algorithms in Control of Machine Processes. Energies 2022, 15, 1895.

• Naosekpam, V. and Sahu, N., 2022, April. IFVSNet: Intermediate Features Fusion based CNN for Video Subtitles Identification. In 2022 IEEE 7th International conference for Convergence in Technology (I2CT) (pp. 1-6). IEEE.

• Boga, J. and Kumar, V.D., 2022. Human activity recognition by wireless body area networks through multi‐objective feature selection with deep learning. Expert Systems, p.e12988.

• Župerl, U., Stepien, K., Munđar, G. and Kovačič, M., 2022. A Cloud-Based System for the Optical Monitoring of Tool Conditions during Milling through the Detection of Chip Surface Size and Identification of Cutting Force Trends. Processes, 10(4), p.671.

• Neto, J.B.C., Ferrari, C., Marana, A.N., Berretti, S. and Bimbo, A.D., 2022. Learning Streamed Attention Network from Descriptor Images for Cross-resolution 3D Face Recognition. ACM Transactions on Multimedia Computing, Communications, and Applications (TOMM).

• Chattopadhyay, S., Dey, A., Singh, P.K., Ahmadian, A. and Sarkar, R., 2022. A feature selection model for speech emotion recognition using clustering-based population generation with hybrid of equilibrium optimizer and atom search optimization algorithm. Multimedia Tools and Applications, pp.1-34.

• Kanagaraj, R., Elakiya, E., Rajkumar, N., Srinivasan, K. and Sriram, S., 2022, January. Fuzzy Neural Network Classification Model for Multi Labeled Electricity Consumption Data Set. In 2022 4th International Conference on Smart Systems and Inventive Technology (ICSSIT) (pp. 1037-1041). IEEE.

AI Glossary - What Is The ARTMAP-IC?

 

What Is The ARTMAP-IC Algorithm?

The fundamental fuzzy ARTMAP is enhanced by this network with distributed prediction and category instance counting.

How Is The ARTMAP-IC Used For Medical Diagnosis?

Medical diagnosis with ARTMAP-IC: Inconsistent cases and instance counting. 

The ARTMAP-IC neural network extends the fundamental fuzzy ARTMAP system with distributed prediction and category instance counting for challenging database prediction issues like medical diagnosis. 

A new version of the ARTMAP match tracking algorithm, which governs search after a predictive mistake, makes prediction with sparse or inconsistent data easier. 

The new approach (MT-) significantly compresses memory without sacrificing speed while improving the accuracy of the real-time network differential equations as compared to the old match tracking algorithm (MT+). 

Simulated analyses of four medical databases—Pima Indian diabetes, breast cancer, heart disease, and gallbladder removal—examine the prognostic accuracy of these conditions. 

Results using logistic regression, K closest neighbor (KNN), the ADAP preceptron, multisurface pattern separation, CLASSIT, instance-based (IBL), and C4 are comparable to or superior to those from ARTMAP-IC. 

The dynamics of ARTMAP are quick, reliable, and scalable. 

By repeatedly training the system on various input set orderings, a voting technique enhances prediction. 

Confidence intervals for competing predictions are derived from voting, instance counting, and distributed representations.

HOW DOES ARTMAP-IC NEURAL NETWORK CLASSIFIER FUNCTION?

In an ART-based network, information reverberates between the network’s layers. 

Learning is possible in the network, when resonance of the neuronal activity occurs. ART1 was developed to perform clustering on binary-valued patterns. 

By interconnecting two ART1 modules, ARTMAP was the first ART-based architecture suited for classification tasks. 

ARTMAP- IC adds to the basic ARTMAP system new capabilities designed to solve the problem with inconsistent cases, which arises in prediction, where similar input vectors correspond to cases with different outcomes, (Carpenter, Grossberg, and Reynolds, 1991), (Carpenter and Markuzon, 1998). 

It modifies the ARTMAP search algorithm to allow the network to encode inconsistent cases (IC). 

Below figure, adapted from (Carpenter and Markuzon, 1998), shows the architecture of an ARTMAP-IC network. 

Simplified ARTMAP-IC Architecture

It consist of fully connected layers of nodes: an M-node input layer F1, an Nnode competitive layer F2, an N-node instance counting layer F3, an L-node output layer F0 b , and an L-node map field Fab that links F3 and F0 b . 

In ARTMAP-IC an input a=(a1, a2, … , aM) learns to predict an outcome b=(b1, b2, …, bL), , where only one component bK=1, placing the input a in class K. 

With fast learning, β=1, ARTMAP-IC represents category K as hyper-rectangle ℜK that just encloses all the training set patterns a to which it has been assigned. 

A set of real weights W={wji: j=1,…,N; i=1,…,M} is associated with the F1 - F2 layer connections. Each F2 node j represents a category in the input space, and stores a prototype vector wj=(wj1, wj2, …,wjM). 

The F2 layer is connected, through associative links to F3, which in turn is connected to the map field Fab by associative links with binary weights Wab=(wjk ab:j=1,…,N; k=1,…,L}. 

The vector wj ab=(wj1 ab, wj2 ab, …,wjL ab) relates F2 node j to one of the L output classes. Instance counting biases distributed predictions according to the number of training set inputs classified by each F2 node. 

During testing the F2->F3 input yj is multiplied by the counting weight cj to produce normalized F3 activity, which projects to the map field Fab for prediction. 

How Does The ARTMAP-IC Algorithm Operate In Classifier Mode?

The following algorithm describes the operation of an ARTMAP-IC classifier in learning mode: 

1. Initialization: 

Initially, all the neurons of F2 are uncommitted, all weight values wji are initialized to 1, and all weight values wjk of Fab are set to 0. 

2. Input pattern coding: 

When a training pair (a,b) is presented to the network, a undergoes preprocessing, and yields pattern A=(A1,A2,…,A2M). 

The vigilance parameter ρ is reset to its baseline value. 

3. Prototype selection: 

Pattern A activates layer F1 and is propagated through weighted connections W to layer F2. 

Activation of each node j in the F2 layer is determined by the choice function Tj(A)=|A∧wj|/(α+|wj|). 

The F2 layer produces a winner-take-all pattern of activity y=(y1,y2,…,yN) such that only node j=J with the greatest activation value remains active (yJ=1). 

Node J propagates its prototype vector wJ back onto F1 and the vigilance test |A∧wj|≥ρM is performed. 

This test compares the degree of match between wJ and A to the vigilance parameter ρ∈[0,1]. 

If this test is satisfied, node J remains active and resonance is said to occur. 

Otherwise, the network inhibits the active F2 node and searches for another node J that passes the vigilance test. 

If such a node does not exist, an uncommitted F2 node becomes active and undergoes learning (step 5). 

4. Class prediction: 

Pattern b is fed directly to the map field Fab, while the F2 activity pattern y is propagated to the map field via associative connections Wab. 

The latter input activates Fab nodes according to the prediction function ∑= = N j ab j jk ab Sk y y w 1 ( ) and the most active Fab node K yields the class prediction (K=k(J)). 

If node K constitutes an incorrect class prediction, a match tracking signal raises vigilance just enough to induce another search among F2 nodes (step 3). 

This search continues until either an uncommitted F2 node becomes active (learning ensues at step 5), or a node J that has  previously learned the correct class prediction K becomes active. 

5. Learning: 

Learning input a involves updating prototype vector wJ, and if J corresponds to a newly committed node, creating a permanent associative link to Fab. 

A new association between F2 node J and Fab node K (K=k(J)) is learned by setting wJk ab=1 for k=K, where K is the target class label for a. 

Once the weights (W and Wab) have converged for the training set patterns, ARTMAP can predict a class label for an input pattern by performing steps 2, 3 and 4 without any testing. 

A pattern a that activates node J is predicted to belong to the class K=k(J)

~ Jai Krishna Ponnappan

Find Jai on Twitter | LinkedIn | Instagram

Be sure to refer to the complete & active AI Terms Glossary here.

You may also want to read more about Artificial Intelligence here.

Reference And Further Reading

• Tayyebi, S. and Soltanali, S., Fuzzy Modeling System Based on Ga Fuzzy Rule Extraction and Hybrid of Differential Evolution and Tabu Search Approaches: Application in Synthesis Gas Conversion to Valuable Hydrocarbons Process. Saeed, Fuzzy Modeling System Based on Ga Fuzzy Rule Extraction and Hybrid of Differential Evolution and Tabu Search Approaches: Application in Synthesis Gas Conversion to Valuable Hydrocarbons Process.

• Tang, Y., Qiu, J. and Gao, M., 2022. Fuzzy Medical Computer Vision Image Restoration and Visual Application. Computational and Mathematical Methods in Medicine, 2022.

• Dymora, P., Mazurek, M. and Bomba, S., 2022. A Comparative Analysis of Selected Predictive Algorithms in Control of Machine Processes. Energies 2022, 15, 1895.

• Naosekpam, V. and Sahu, N., 2022, April. IFVSNet: Intermediate Features Fusion based CNN for Video Subtitles Identification. In 2022 IEEE 7th International conference for Convergence in Technology (I2CT) (pp. 1-6). IEEE.

• Boga, J. and Kumar, V.D., 2022. Human activity recognition by wireless body area networks through multi‐objective feature selection with deep learning. Expert Systems, p.e12988.

• Župerl, U., Stepien, K., Munđar, G. and Kovačič, M., 2022. A Cloud-Based System for the Optical Monitoring of Tool Conditions during Milling through the Detection of Chip Surface Size and Identification of Cutting Force Trends. Processes, 10(4), p.671.

• Neto, J.B.C., Ferrari, C., Marana, A.N., Berretti, S. and Bimbo, A.D., 2022. Learning Streamed Attention Network from Descriptor Images for Cross-resolution 3D Face Recognition. ACM Transactions on Multimedia Computing, Communications, and Applications (TOMM).

• Chattopadhyay, S., Dey, A., Singh, P.K., Ahmadian, A. and Sarkar, R., 2022. A feature selection model for speech emotion recognition using clustering-based population generation with hybrid of equilibrium optimizer and atom search optimization algorithm. Multimedia Tools and Applications, pp.1-34.

• Kanagaraj, R., Elakiya, E., Rajkumar, N., Srinivasan, K. and Sriram, S., 2022, January. Fuzzy Neural Network Classification Model for Multi Labeled Electricity Consumption Data Set. In 2022 4th International Conference on Smart Systems and Inventive Technology (ICSSIT) (pp. 1037-1041). IEEE.

AI Glossary - What Is ARTMAP?

 

What Is ARTMAP AI Algorithm?

The supervised learning variant of the ART-1 model is ARTMAP.

It learns binary input patterns that are given to it.

The suffix "MAP" is used in the names of numerous supervised ART algorithms, such as Fuzzy ARTMAP.

Both the inputs and the targets are clustered in these algorithms, and the two sets of clusters are linked.

The ARTMAP algorithms' fundamental flaw is that they lack a way to prevent overfitting, hence they should not be utilized with noisy data.

How Does The ARTMAP Neural Network Work?

A novel neural network architecture called ARTMAP automatically picks out recognition categories for any numbers of arbitrarily ordered vectors depending on the accuracy of predictions. 

A pair of Adaptive Resonance Theory modules (ARTa and ARTb) that may self-organize stable recognition categories in response to random input pattern sequences make up this supervised learning system. 

The ARTa module gets a stream of input patterns ([a(p)]) and the ARTb module receives a stream of input patterns ([b(p)]), where b(p) is the right prediction given a (p). 

An internal controller and an associative learning network connect these ART components to provide real-time autonomous system functioning. 

The remaining patterns a(p) are shown during test trials without b(p), and their predictions at ARTb are contrasted with b. (p). 

The ARTMAP system learns orders of magnitude more quickly, efficiently, and accurately than alternative algorithms when tested on a benchmark machine learning database in both on-line and off-line simulations, and achieves 100% accuracy after training on less than half the input patterns in the database. 

It accomplishes these features by using an internal controller that, on a trial-by-trial basis, links predictive success to category size and simultaneously optimizes predictive generalization and reduces predictive error, using only local operations. 

By the smallest amount required to rectify a predicted inaccuracy at ARTb, this calculation raises the alertness parameter an of ARTa. 

To accept a category or hypothesis triggered by an input a(p), rather than seeking a better one via an autonomously controlled process of hypothesis testing, ARTa must have a minimal level of confidence, which is calibrated by the parameter a. 

The degree of agreement between parameter a and the top-down learnt expectation, or prototype, which is read out after activating an ARTa category, is compared. 

If the degree of match is less than a, search is initiated. 

The self-organizing expert system known as ARTMAP adjusts the selectivity of its hypotheses depending on the accuracy of its predictions. 

As a result, even if they are identical to frequent occurrences with distinct outcomes, unusual but significant events may be promptly and clearly differentiated. 

In the intervals between input trials, a returns to baseline alertness. 

When is big, the system operates in a cautious mode and only makes predictions when it is certain of the result. 

At no step of learning, therefore, do many false-alarm mistakes happen, yet the system nonetheless achieves asymptote quickly. 

Due to the self-stabilizing nature of ARTMAP learning, it may continue to learn one or more databases without deteriorating its corpus of memories until all available memory has been used.

What Is Fuzzy ARTMAP?

For incremental supervised learning of recognition categories and multidimensional maps in response to arbitrary sequences of analogue or binary input vectors, which may represent fuzzily or crisply defined sets of characteristics, a neural network architecture is developed. 

By taking advantage of a close formal resemblance between the computations of fuzzy subsethood and ART category choosing, resonance, and learning, the architecture, dubbed fuzzy ARTMAP, accomplishes a synthesis of fuzzy logic and adaptive resonance theory (ART) neural networks. 

In comparison to benchmark backpropagation and general algorithm systems, fuzzy ARTMAP performance was shown using four simulation classes. 

A letter recognition database, learning to distinguish between two spirals, identifying locations inside and outside of a circle, and incremental approximation of a piecewise-continuous function are some of the simulations included in this list. 

Additionally, the fuzzy ARTMAP system is contrasted with Simpson's FMMC system and Salzberg's NGE systems.

~ Jai Krishna Ponnappan

Find Jai on Twitter | LinkedIn | Instagram

Be sure to refer to the complete & active AI Terms Glossary here.

You may also want to read more about Artificial Intelligence here.

References And Further Reading:

• Moreira-Júnior, J.R., Abreu, T., Minussi, C.R. and Lopes, M.L., 2022. Using Aggregated Electrical Loads for the Multinodal Load Forecasting. Journal of Control, Automation and Electrical Systems, pp.1-9.

• Ferreira, W.D.A.P., Grout, I. and da Silva, A.C.R., 2022, March. Application of a Fuzzy ARTMAP Neural Network for Indoor Air Quality Prediction. In 2022 International Electrical Engineering Congress (iEECON) (pp. 1-4). IEEE.

• La Marca, A.F., Lopes, R.D.S., Lotufo, A.D.P., Bartholomeu, D.C. and Minussi, C.R., 2022. BepFAMN: A Method for Linear B-Cell Epitope Predictions Based on Fuzzy-ARTMAP Artificial Neural Network. Sensors, 22(11), p.4027.

• Santos-Junior, C.R., Abreu, T., Lopes, M.L. and Lotufo, A.D., 2021. A new approach to online training for the Fuzzy ARTMAP artificial neural network. Applied Soft Computing, 113, p.107936.

• Ferreira, W.D.A.P., 2021. Rede neural ARTMAP fuzzy implementada em hardware aplicada na previsão da qualidade do ar em ambiente interno.