Science topics: Artificial IntelligenceNeural Networks (Computer)

Science topic

# Neural Networks (Computer) - Science topic

Explore the latest questions and answers in Neural Networks (Computer), and find Neural Networks (Computer) experts.

Questions related to Neural Networks (Computer)

Our study goal is to develop and validate questionnaire for patient satisfaction and then using this questionnaire tool to evaluate patient satisfaction then building a model to predict patient satisfaction using artificial neural network. need help to calculate sample size for this study.

In my opinion, the perceptron and other machine learning algorithms can evaluate quite complex functional dependencies of time series, if you have any ideas for further research in this vein, welcome to a private or public discussion.

The study presents a bio-inspired chaos sensor model based on the perceptron neural network for the estimation of entropy of spike train in neurodynamic systems.

how to implement neural network for 2D planar robotic manipulator to estimate joint angles for a commanded position in a circular path? and how to estimate its error for defined mathematical model and neural network model in a circular path??

What type of deep learning architectures should we prefer while working on CNN models, Standards models such as AlexNET, VGGNET, or Customised models (with user-defined layers in the neural network architecture)?

In my research, I need to compare the neural networks I have built, consisting mainly of perceptron and normalization layers, with networks from other publications that have convolution, pulling, normalization and perceptron layers in terms of computational complexity. I have the ability to calculate the number of parameters a given neural network has on a given layer, but I don't know how I should compare it.

Should I take only convolution layers as the most taxing, or sum the number of parameters from all of layers?

How should I compare neural networks that have computationally stressful convolution layers with others that do not have them, but perform feature extraction in a different way?

I noticed that in some very bad models of neural networks, the value of R² (coefficient of determination) can be negative. That is, the model is so bad that the mean of the data is better than the model.

In linear regression models, the multiple correlation coefficient (R) can be calculated using the root of R². However, this is not possible for a model of neural networks that presents a negative R². In that case, is R mathematically undefined?

I tried calculating the correlation y and y_pred (Pearson), but it is mathematically undefined (division by zero). I am attaching the values.

Obs.: The question is about artificial neural networks.

I am building a Neural Network Model for approximating SINR value at a specified location in 2D. The scenario consists of a central gNB placed at (1500, 1500), and other 6 gNBs are placed surrounding it in a hexagonal manner with ISD equal to 1000 m. The UEs are placed around the central gNB at distances from 30m to 500m in steps of 20m, along X and Y. I thus have 2500 points at which I am getting SINR.

For training, X and Y coordinates of UE and their values obtained from NetSim simulation have been used. Thus, the features are coordinates (X, Y), while the Label is SINR.

The NN model architectures I have tried using (i) 2-5 layers (ii) Activation functions: ReLU, LeakyReLU, tanh, sigmoid

(iii) Different units size: 32 to 128, (iv) Dropout at various layers from 0.4 to 0.5,

However, the minimum MSE obtained was around 37 while training, which is very high.

I have trained a deep neural network model. Now I want to use information from this model to predict. One way is to load the model and then predict. But is there any way we predict without loading or training the model?

I want to write a simple python code in which I manually give data from the trained model and do predictions without using any library. But I don't know how and what type of information should I extract from the model?

I carried out few experiments by combining 7 CNN based feature extractors and 7 machine learning classifier. In total there were 49 pairs.

For CNNs i used VGG16, Mobile net v2, Densenet121, Inception V3, ResNet 101, ResNet 152 and XceptionNet as feature extractors and passed the generated feature vector to ML classifiers to perform binary classification.

The ML classifiers i have used are support vector machine, k nearest neighbour, gaussian naive Bayes, decision tree, random forest, extra tress and Multi layer perceptron.

For all the evaluation metrics like accuracy, precision, recall and F1 score i achieved best results with the combination of ResNet 101 and Multi layer perceptron.

I'm not able to understand that why it is performing the best. Resnet152 has a deeper network and support vector machine generally perform well. In my case Resnet101 and multi layer perceptron is giving the best results.

Please help me to understand the reason behind it.

I have come across many research articles stating the values of rmse of deep learning models such as LSTM. Some results are between 0 and 1, while some are between 0 and 50. I want to know what an ideal rmse value is for such neural network models.

I have been trying to reduce the rmse values to less than 1 from values between 3 and 7.

Note: The error values are computed after unscalling back the dataset to its original form. The errors are, however, between 0 and 1 when computed with scaled data. But, I think computing with unscaled data makes more sense.

Generalized Regression Neural Network Model is available in nntool in Matlab? if not, then how can I use it?

While training a neural network model for multiclass datasets using

**variables***Keras, earth-analytics, and TensorFlow,***"accuracy"**and**“loss”**per epoch were carried out.Now the questions are:

- How to interpret these variables?
- How do they affect the behavior of the trained model?
- How to define the appropriate
**epochs number**for a better modeling**performance**?

Following is the error faced while training the deep learning CNN model on the dataset. The dataset is of MRI images of prostate cancer in men and have dcm extension. I tried training the model with the dcm extension as well as by converting it to jpg but the error remains the same. Could anyone help me solving this error?

**Code for reference:**

import pydicom as dicom

PNG = False

folder_path = "/tmp/Prostate_dataset/Prostate_dataset-1"

jpg_folder_path = "/tmp/prostate_jpg"

images_path = os.listdir(folder_path)

for n, image in enumerate(images_path):

ds = dicom.dcmread(os.path.join(folder_path, image))

pixel_array_numpy = ds.pixel_array

if PNG == False:

image = image.replace('.dcm', '.jpg')

else:

image = image.replace('.dcm', '.png')

cv2.imwrite(os.path.join(jpg_folder_path, image), pixel_array_numpy)

if n % 50 == 0:

print('{} image converted'.format(n))

def get_data(path):

data = []

for img in os.listdir(path):

img_arr=os.path.join(path,img)

data_img=cv2.imread(img_arr)

data.append(data_img)

return np.array(data)

train = get_data(jpg_folder_path)

val = get_data(jpg_folder_path)

# Normalize the data

train = np.asarray(train)

train = train.reshape(-1, 256,256,1)

train = train.astype('float64')

train /= 255.0

val = np.asarray(val)

val = val.reshape(-1, 256,256,1)

val = val.astype('float64')

val /= 255.0

model = tf.keras.models.Sequential([

tf.keras.layers.Conv2D(128, (5,5), activation='relu', input_shape=(256,256,1)),

tf.keras.layers.BatchNormalization(axis=-1),

tf.keras.layers.Dropout(.5, noise_shape=None, seed=None),

tf.keras.layers.MaxPooling2D(2, 2),

tf.keras.layers.Conv2D(64, (5,5), activation='relu'),

tf.keras.layers.BatchNormalization(axis=1),

tf.keras.layers.Dropout(.5, noise_shape=None, seed=None),

tf.keras.layers.MaxPooling2D(2,2),

tf.keras.layers.Conv2D(16, (5,5), activation='relu'),

tf.keras.layers.BatchNormalization(axis=1),

tf.keras.layers.Dropout(.5, noise_shape=None, seed=None),

tf.keras.layers.MaxPooling2D(2,2),

tf.keras.layers.Conv2D(8, (5,5), activation='relu'),

tf.keras.layers.BatchNormalization(axis=1),

tf.keras.layers.Dropout(.5, noise_shape=None, seed=None),

tf.keras.layers.MaxPooling2D(2,2),

tf.keras.layers.Flatten(),

tf.keras.layers.Dense(128, activation='relu'),

tf.keras.layers.Dense(256, activation='softmax'),

])

model.summary()

model.compile(optimizer='adam',loss='categorical_crossentropy',metrics=['accuracy'])

model.fit(X_train, y_train, epochs = 10, verbose = 1, validation_data = (X_test, y_test))

test_loss = model.evaluate(X_test, y_test)

**Error:**

ValueError: in user code: /usr/local/lib/python3.7/dist-packages/keras/engine/training.py:853 train_function * return step_function(self, iterator) /usr/local/lib/python3.7/dist-packages/keras/engine/training.py:842 step_function ** outputs = model.distribute_strategy.run(run_step, args=(data,)) /usr/local/lib/python3.7/dist-packages/tensorflow/python/distribute/distribute_lib.py:1286 run return self._extended.call_for_each_replica(fn, args=args, kwargs=kwargs) /usr/local/lib/python3.7/dist-packages/tensorflow/python/distribute/distribute_lib.py:2849 call_for_each_replica return self._call_for_each_replica(fn, args, kwargs) /usr/local/lib/python3.7/dist-packages/tensorflow/python/distribute/distribute_lib.py:3632 _call_for_each_replica return fn(*args, **kwargs) /usr/local/lib/python3.7/dist-packages/keras/engine/training.py:835 run_step ** outputs = model.train_step(data) /usr/local/lib/python3.7/dist-packages/keras/engine/training.py:789 train_step y, y_pred, sample_weight, regularization_losses=self.losses) /usr/local/lib/python3.7/dist-packages/keras/engine/compile_utils.py:201 __call__ loss_value = loss_obj(y_t, y_p, sample_weight=sw) /usr/local/lib/python3.7/dist-packages/keras/losses.py:141 __call__ losses = call_fn(y_true, y_pred) /usr/local/lib/python3.7/dist-packages/keras/losses.py:245 call ** return ag_fn(y_true, y_pred, **self._fn_kwargs) /usr/local/lib/python3.7/dist-packages/tensorflow/python/util/dispatch.py:206 wrapper return target(*args, **kwargs) /usr/local/lib/python3.7/dist-packages/keras/losses.py:1666 categorical_crossentropy y_true, y_pred, from_logits=from_logits, axis=axis) /usr/local/lib/python3.7/dist-packages/tensorflow/python/util/dispatch.py:206 wrapper return target(*args, **kwargs) /usr/local/lib/python3.7/dist-packages/keras/backend.py:4839 categorical_crossentropy target.shape.assert_is_compatible_with(output.shape) /usr/local/lib/python3.7/dist-packages/tensorflow/python/framework/tensor_shape.py:1161 assert_is_compatible_with raise ValueError("Shapes %s and %s are incompatible" % (self, other))

ValueError: Shapes (None, 256, 256, 1) and (None, 256) are incompatible

One of famous method in Machine Learning is Perceptron Rule. In following Book and it's R Perceptron function some of problems are not working correctly. I develop R code that worked with several examples and exercises.

# R Deep Learning Project, Y. Liu, P. Maldonado

One of simple problem is showing in this picture that separate points in two groups with correctly decision boundary line.

While modeling Lynx series after applying log10 transformation, I get ARIMA(2,0,3) as the best fitted model using auto.arima in R. But I see Zhang 2003 (Time series forecasting using a hybrid ARIMA and neural network model, G. Peter Zhang) specifies, after applying log10 transformation, ARIMA(12,0,0) as the best fitted model using MATLAB. This has also been followed up by many researchers Khandelwal et al. 2015, Adhikari R and Agrawal R.K 2011, etc. Can some one clarify if R and MATLAB can provide different models as the best fitted model(s)?

r2 for MLR = 10-20 % approx at various train:test ratio

but for ANN it is 1-4%.

why?

thanks in advance

I am using nonlinear autoregressive neural network model (NAR-NN) to predict future values from historical record. I made the model using the ntstool in MATLAB but I couldn't apply & implement the model to predict future values. So, how to implement the model for future data prediction?

What should consider to compare performance of two neural network models? What are the parameters required to increase model accuracy?

I want to use the General regression neural network(GRNN) model in r-software to simulate the reactions of aboveground biomasses of perennial trees to climate change factors (rainfall, minimum and maximum temperature). However, I felt compelled to utilize the program to install the GRNN package, preparing the dataset, and execute the model.

Is there anyone who can help me with how to install the GRNN package and executing the R model?

Hi, Is there any way to include the tensorflow lite on the Teensy board? I have been searching online but it seems that CMSIS NN is the only way to run neural network model on Teensyduino. I have tried to include the Tensorflow lite library but it won't run. Thanks,

As I have observed, if we increase window size then it increases accuracy which in turn slowdowns the training. Contrarily, if we decrease input window size then accuracy decreases, and the network gets trained faster. Need suggestions for obtaining the best results from LSTM network with optimal window size.

Hi All,

I have recently started working on neural networks for hardware implementation where we normally work with fully connected networks for pattern classification. I am facing this issue as to how to report the pattern classification accuracy. There are 2 situations that mainly arises in this context:

1) How to report the training accuracy vs epoch data? whether this should be some sort of mean data over several runs for every epoch or the best performing data?

2) How to report the test accuracy for the trained model?

NB: There has been some discussions on stack exchange on this (https://stats.stackexchange.com/questions/205610/reporting-of-neural-network-accuracy-for-academic-publications) but its not very conclusive

hi dears!

i have 4 inputs and one output neural network model. i want to extract the optimum input from the four inputs. some one who has MATLAB code, please share me.

thank you!

Dear researchers

**Objective:**We've applied machine learning methods such as artificial neural networks, random forest, and support vector machines to predict stroke patient's recovery.

**Materials and methods:**We have stroke patients' clinical data from EMRs(electronic medical records) and their kinematic data obtained by the exoskeleton robot's sensor system(from gait training).

The clinical data are ordinal and categorical, and the kinematic data are time-series data.

Clinical data and kinematic data have been integrated into tabular data by applying moving windows to time-series data (obtained mean, std, median, max, and min).

**Limitations:**In our experience, it was not easy to use all the data for training at once because the types and characteristics of clinical data and kinematic data were different.

Thus, we are applying the ensembling method to various neural network models.

(We've tried conventional bagging or stacking algorithms to the outputs of the neural networks.)

**Question:****At this point, we would like to know some reasonable, preferred, recommended methods for ensembling the neural network models with different data learned separately.**(i.e., how to combine a neural network model trained by clinical data and another model trained by kinematic data)

The WuDao 2.0 model of China Neural Network (NN) with 1,75 trillion parameters topped the 1.6 trillion that Google unveiled in a similar model in January 2021. Is it the start of Race to Quadrillion parameters NN? Do you have additional information about the structure and design of such ultra big NN?

In order to make a comparison between experimental results and results obtained by machine learning models or regressions . I'd like to know how is it possible to predict results of (just ONE output ) according to the FOUR inputs illustrated in the attached picture using NEURAL NETWORKS using PYTHON where all data are numbers? and which is the best NEURAL NETWORK MODEL that gives the best prediction in this case ?

The best simulator that can aid the design of a Bayesian Neural Network model.

According to the nature of temporal data, strategies like k-fold cross validation is not an appropriate idea since we cannot remove the dimension of time. In this discussion we want to explore ideas about testing models for temporal data.

I'm learning how to use the Recurrent Neural Network model (RNN). I'm not entirely sure about the feed-forward procedure in RNN. It includes, for example, input, hidden state, and output. As far as I know, the hidden state is a type of multi-layer perception (MLP). However, in this case, a hidden state is derived from both current input and a previously hidden state. Unfortunately, "we can note that everyone reported the total number of memory cells, but no one specified the number of neurons inside each memory cell," which confuse me.

Second, I am confused by the RNN backpropagation procedure. I searched Google, and everyone only mentioned the generic steps (calculations of gradients) but no one conducted

**step-by-step backpropagation****on example**. I'm desperate for the entire RNN training process (two iterations is sufficient), even on a single layer with three to four memory cells.Can anyone available concrete examples of the RNN model training process?

I am working on a novel image classification problem with only 128 images of one class and 25 images of another class. From CNN perspective, I have too little data to learn features well and on top of that, the dataset is highly imbalanced. What strategies should I adopt to improve the model in this scenario? Could other techniques could be tried (SVM ?)

Why in simulink neural network model reference control example the inputs to the controller consist of two delayed reference inputs, two delayed plant outputs, and one delayed controller output? Also why the plant has 2 inputs?

In RNN, Y

_{t}is the output which is given by the equation;**Y**

_{t }= W_{hy}.h_{t }where W

_{hy }is the weight associated with the output layer and h_{t}is the current state.What is the need of having this weight at the output layer of an RNN? Is it anything

**related to the memory**of the network?Sir ,

When there is a fault on the lower H-BRIDGE diagonal switches the waveforms of output voltage is similar . How did the neural network model classify the fault in this case ?

Why most researchers are shifting from tensorFlow to Pytorch?

Hi, I am working on an Industrial Process "Tennessee Eastman Process" and I need to train Neural Network model for its reactor part using BAT optimisation algorithm in MATLAB/Simulink.

MATLAB neural network app doesn't provide the option of selecting BAT learning algorithm while training the model.

Can anyone guide me or help me in finding the code that can show how to train ANN model using BAT algorithm?

I have attached the relevant paper which uses BAT algorithm to train ANN model.

Performance forecasting...

I want to start the training of my ANN with 13 input data, three layers. my problem is, does my actual output has to be 13 input data as well? or can it be any number too.

How do i bring down the different unit of measurements? i have data with units of measurements in; hours, days, percentage, tons, TEU, metres, and monetary units. to make it simple, data representation.?

I'm a hydrologist and pretty new to machine learning, but would like to use sensor data that I have (rainfall time series for example) and combine it with GIS data (grids/rasters of topography for example) as the input to a neural network to then produce a variable of interest (streamflow for example). I can easily take a 1d-array of daily rainfall and match it up with a 1d-array of stream flow and set up and train a regression perceptron (vanilla) network in Keras, but I am having a hard time wrapping my head around how it would be best to combine the single timeseries values with static but spatially distributed topography data to create a single input array of input/training data to then ultimately try to predict the 1d-array of streamflow data.

Any assistance regarding how to format these data sets would be appreciated. Thanks!

I'm working on a Project to improve the vehicle Position accuracy based on vehicle data. Basically I'm trying to make an Algorithm that can estimate the vehicle position with neural networks where my input data are the vehicle data like velocity, acceleration, braking pressure etc.. and my outputs are latitude and longitude. I''m doing a research about this but I didn't find any good papers about this topic. there is some good related papers but they are about using Kalman Filters or INS data but my Goal is to achieve this with using vehicle data from sensors like I said, for example velocity, acceleration, Braking, steer wheel, current Gear etc.. is there some good refences about this? I hope someone here have Experience about this and can help me with some advices

If a deep neural network model (Net A) produces higher accuracy than another deep net model (Net B) for a given dataset, will Net A always produce better accuracy than Net B for any dataset given that all other parameters are constant?

I have a multivariate time-series forecasting problem which I am trying to solve using LSTM models. I have transformed my data such that it looks back last 20 timestamps to reconstruct the records at current timestamp, using deep LSTM model. Following is the model architecture :

_{model3 = Sequential()}

_{model3.add(LSTM(1000,input_shape=(train_X.shape[1], train_X.shape[2]), return_sequences = True))}

_{model3.add(Dropout(0.2))}

_{model3.add(LSTM(500, activation = 'relu', return_sequences = True))}

_{model3.add(Dropout(0.2))}

_{model3.add(LSTM(250, activation = 'relu', return_sequences = False))}

_{model3.add(Dropout(0.2))}

_{model3.add(Dense(train_y.shape[1]))}

_{model3.compile(loss='mse', optimizer='adam', metrics = ['accuracy'])}

_{model3.summary()}

^{history3 = model3.fit(train_X, train_y, epochs=100, batch_size=36, validation_data=(test_X, test_y), verbose = 2, shuffle= False)}

^{Attached is the graph of the neural network output. The 'validation loss' metrics from the test data has been oscillating a lot after epochs but not really decreasing. }

^{Can anyone explain how to interpret the graph ? Also, what could be the potential ways to ensure that the model does get improved with every new epoch ?}

**Cross-validation**, it’s a model validation techniques for assessing how the results of a statistical analysis (model) will generalize to an independent data set

If any such a related research article or text books are available kindly list it..

I am training a neural network model in R but the solutions are not converging during training. so as a result i am not getting the desired output during prediction (r2 coming below 1). Can anyone tell me how can I get a better training model? and if anyone can provide me a modified code of neural net in R?

Hai,

I have dataset with 100s of independent features (column wise) and one column of dependent (categorical like good or bad). I would like to develop a neural network model which train on this dataset.

in this below mentioned matlab code, net = feedforwardnet(10); net = train(net,x,t); view(net) y = net(x); perf = perform(net,y,t)

x = (6500x100)double

t = (6500x1)categorical (good/bad)

Neural network training is not working. why? and how can I fix this problem?

is it okay to get data from ansys or abaqus and build a model of neural network Or I must get the data from a realistic field experiment ?

hi, I would like to calculate the distance of a step with an accelerometer 6 axis 3 accelero 3 gyro attach to the belt of the human . When I integrate, I'm losing too much accuracy.

I see some research about a way to determine the distance of steeps with neural network models. Does anyone have an idea if it is really possible and is this method precise ?

If yes, do someone know the type of the model and the input parameters ?

If someone has another idea of who could I have the distance of one step with an accelerometer ?

Thanks

Hello everyone.

I am dealing with construction of extrapolation model with deep learning.

When the training dataset is a shape of (1,0,0,0), (0,1,0,0), (0,0,1,0), (0,0,0,1)

I want to build up a model that is able to reconstruct data that is shape of (1,1,0,0 ), (1,1,1,0).

Since the training data and test (or actually-using) data is different, it does not make sense for conventional facts related to deep learning.

The task now I am doing is, using the test data as a validation set of learning and finding the learning that minimize validation loss.

However, now I realize that the task costs too much resource and time. It can be same as doing some lottery.

Therefore, there should be some improvement for the methods.

If someone can give me some advice for this part, it would be a really big help.

Thank you so much.

Hi

I am currently trying to interpret the predictions of a fully connected Neural Network on a regression task. The inputs are adjacency matrices for a brain neural networks (from diffusion MRI scans) to predict cognitive abilities. My model currently does a good job at predicting (mean error of 5%) - I would like to understand why it is making the decisions it is making (e.g. which regions of the brain are more important).

I have come across a lot of literature covering this problem, but only regarding constitutional neural networks, not fully connected ones.

Thanks!

The achievement of high-quality results of deep learning is possible by appropriately setting of its parameters. The number of hidden layers and the number of neurons in each layer of a deep machine learning have main influence on the performance of the algorithm. Some manual parameter setting and grid search approaches somewhat ease the users' tasks in setting these important parameters. But, these two techniques can be very time-consuming.

What is the most recent algorithm of soft computing?

I want to test MLP Neural network using leave one out cross validation, is there some approach to obtaining cross-validated R2 using in Matlab?

Best regards

I am doing thesis on baby cry prediction.

i had dataset of baby cries and non- baby cries of two classes.

i want use Mfcc feature extraction technique to identify important components of audio signal and train a model using this feature.

my question is how mfcc knows to select important components of signal.

i got a mfcc plot of baby cry.

can any one explain this plot?

I want to know if I can consider an ELM NN as a Deep Learning approach even if it is a single layer neural network, and in Deep Learning we work with Multi Layer perceptrons.

Thanks in advance

Hope you help me for the algorithm for the bayesian regularization Thank you

i want to detect baby cry in real-time and i am using windows system. I had extracted the mfcc features of baby cry and trained a neural network model.

i got model accuracy 99%. This model i export to raspberry pi for real-time testing and got good results. But the problem is lot of false positives.

the model is detecting every loud sounds it shows lot of false positives.

my aim is to detect only baby cry.

i also used pitch features but no changes in reducing false positives.

please help me to overcome this problem.

I am working on computer vision problem, we have a very images to train. I wanted to know is there any possible and practical way to generate synthetic images using deep learning.

Any suggestions are appreciated.

I am having problem with prediction with my model trained on resnet50. I have 10 classes of Nepali numbers from (0 ...9). I have trained the model for 100 epochs with around 40,000 data . What is the issue with my model? I am having overfit? Or, the model I have used to train is just too complex for 10 class. I have also tried this model to predict on the training set but the prediction accuracy is very very poor around (10%).