I have tried everything to allocate a struct containing a double pointer to device memory using cudaMalloc. I know this question has been asked multiple times yet the question is always for only a single pointer in a structure. I cannot flatten the data because training.data[] index is linked to many other functions of the program. If i cant index the data using training.data[1][n], training.data[2][n]……etc then i cant make use of the data. Ive tried this but keep getting a memory access violation error:

    typedef struct training_data {
   float** data;                    
   float* answer;                   
   float error_rate[500];
}create_training_structure;





create_training_structure *GPUtraining;
float** tempDataSamples;
float* tempData;

cudaStatus = cudaMalloc((void**)&GPUtraining, sizeof(create_training_structure));

if (cudaStatus != cudaSuccess) {
    fprintf(stderr, "cudaMalloc for GPUtraining Structure failed: %sn", cudaGetErrorString(cudaStatus));
}


for (int rows = 0; rows < trainingRows; ++rows) {
    // Allocate memory for data (2D array) on GPU. training[X][]
    cudaStatus = cudaMalloc((void**)&tempDataSamples, trainingRows * sizeof(float*));
    if (cudaStatus != cudaSuccess) {
        fprintf(stderr, "cudaMalloc for GPUtraining[X][] failed: %sn", cudaGetErrorString(cudaStatus));
    }
}

// Allocate memory for data (2D array) on GPU. training[X][]
for (int i = 0; i < trainingRows; ++i) {
    cudaStatus = cudaMalloc((void**)&tempData, (trainingColumns + BIAS_NEURON) * sizeof(float));
    
    if (cudaStatus != cudaSuccess) {
        fprintf(stderr, "cudaMalloc for GPUtraining[][X] failed: %sn", cudaGetErrorString(cudaStatus));
    }
}

for (int rows = 0; rows < trainingRows; ++rows) {
    for (int columns = 0; columns < trainingColumns; columns++) {
        cudaMemcpy(&(GPUtraining->data[rows][columns]), &(training.data[rows][columns]), sizeof(float*), cudaMemcpyHostToDevice);
    }
}