I enjoy making things. Here are a selection of projects that I have worked on over the years.
Agentic AI Systems in Multi-Environment Settings 1. Introduction to Agentic AI An Agentic AI System refers to an autonomous AI system that can sense, decide, and act in an environment to achieve specific goals. In multi-environment settings, these AI agents operate across diverse, dynamic, and often conflicting environments, requiring adaptive decision-making, communication, and coordination.
#include <iostream> #include <vector> #include <png.h> void convertToGrayscale(const std::string& inputFileName, const std::string& outputFileName) { FILE* inputFile = fopen(inputFileName.c_str(), "rb"); if (!inputFile) { std::cerr << "Error opening input image file." << std::endl; return; } png_structp pngPtr = png_create_read_struct(PNG_LIBPNG_VER_STRING, nullptr, nullptr, nullptr); if (!pngPtr) { std::cerr << "Error creating PNG read structure." << std::endl; fclose(inputFile); return; } png_infop infoPtr = png_create_info_struct(pngPtr); if (!infoPtr) { std::cerr << "Error creating PNG info structure." << std::endl; png_destroy_read_struct(&pngPtr, nullptr, nullptr); fclose(inputFile); return; } png_init_io(pngPtr, inputFile); png_read_info(pngPtr, infoPtr); int width = png_get_image_width(pngPtr, infoPtr); int height = png_get_image_height(pngPtr, infoPtr); int bitDepth = png_get_bit_depth(pngPtr, infoPtr); int colorType = png_get_color_type(pngPtr, infoPtr); if (bitDepth != 8 || colorType != PNG_COLOR_TYPE_RGB) { std::cerr << "Unsupported image format. Only 8-bit RGB images are supported." << std::endl; png_destroy_read_struct(&pngPtr, &infoPtr, nullptr); fclose(inputFile); return; } std::vector<unsigned char> imageData(width * height * 3); std::vector<std::vector<unsigned char> > inputImage(height, std::vector<unsigned char>(width * 3)); for (int i = 0; i < height; ++i) { png_read_row(pngPtr, reinterpret_cast<png_bytep>(inputImage[i].data()), nullptr); } fclose(inputFile); png_destroy_read_struct(&pngPtr, &infoPtr, nullptr); // Convert to grayscale std::vector<std::vector<unsigned char> > outputImage(height, std::vector<unsigned char>(width)); for (int i = 0; i < height; ++i) { for (int j = 0; j < width; ++j) { // Simple grayscale conversion formula: (R + G + B) / 3 outputImage[i][j] = (inputImage[i][j * 3] + inputImage[i][j * 3 + 1] + inputImage[i][j * 3 + 2]) / 3; } } // Write output image FILE* outputFile = fopen(outputFileName.c_str(), "wb"); if (!outputFile) { std::cerr << "Error opening output image file." << std::endl; return; } pngPtr = png_create_write_struct(PNG_LIBPNG_VER_STRING, nullptr, nullptr, nullptr); if (!pngPtr) { std::cerr << "Error creating PNG write structure." << std::endl; fclose(outputFile); return; } infoPtr = png_create_info_struct(pngPtr); if (!infoPtr) { std::cerr << "Error creating PNG info structure." << std::endl; png_destroy_write_struct(&pngPtr, nullptr); fclose(outputFile); return; } png_init_io(pngPtr, outputFile); png_set_IHDR(pngPtr, infoPtr, width, height, 8, PNG_COLOR_TYPE_GRAY, PNG_INTERLACE_NONE, PNG_COMPRESSION_TYPE_DEFAULT, PNG_FILTER_TYPE_DEFAULT); png_write_info(pngPtr, infoPtr); for (int i = 0; i < height; ++i) { png_write_row(pngPtr, reinterpret_cast<png_bytep>(outputImage[i].data())); } png_write_end(pngPtr, nullptr); png_destroy_write_struct(&pngPtr, &infoPtr); fclose(outputFile); std::cout << "Image processing complete. Grayscale image saved to " << outputFileName << std::endl; } int main() { std::string inputFileName = "input.png"; std::string outputFileName = "output_grayscale.png"; convertToGrayscale(inputFileName, outputFileName); return 0; }
Using NLP and Language Models to Address Toxicity and Unconscious Bias Theoretical Foundations of NLP for Addressing Toxicity and Bias To rigorously address toxicity and unconscious bias in language models, we rely on linguistic theories, probabilistic models, fairness-aware AI principles, adversarial learning, and information theory. Below, I outline the strong theoretical basis underpinning these techniques.
