How probability, feature engineering, HMMs, GMMs, and evaluation practices shaped machine learning, speech recognition, and computer vision in the 1990s and 2000s.
PRASA and the Early Development of Speech Technology in Southern Africa
How PRASA conference research connected speech recognition, language identification, synthesis, and low-resource language technology to pattern recognition in southern Africa.
Medical Image Analysis Before Deep Learning: Segmentation, Features, and Evaluation
A historical look at pre-deep-learning medical image analysis, including segmentation, hand-engineered features, classifiers, registration, and clinical evaluation.
Stereo Vision in the Mid-2000s: Correspondence, Calibration, and Practical Depth
How mid-2000s stereo vision systems turned disparity into depth, and why calibration, occlusion handling, synchronization, and confidence mattered in practice.
How Mid-2000s Speech Recognition Modeled Uncertainty
How mid-2000s speech recognizers used HMMs, GMMs, n-gram language models, decoding, and adaptation to manage uncertainty in spoken language.
Language Identification in the 2000s: N-Grams, Speech Models, and Uncertainty
How 2000s researchers identified languages in short text and speech using character n-grams, acoustic and phonotactic models, evaluation methods, and rejection thresholds.
Low-Resource Speech Recognition in the 2000s: Data, Design, and Evaluation
How 2000s speech-recognition projects addressed scarce data, multilingual modeling, orthography, code-switching, evaluation, and community governance for low-resource languages.
Visual Tracking in the Early 2000s: Motion, Appearance, and Identity
A historical look at early-2000s visual tracking, from Kalman and particle filters to data association, detection-based tracking, stereo, and evaluation.
Document Analysis in the 2000s: Beyond OCR
How 2000s document-analysis systems handled page layout, reading order, tables, forms, handwriting, historical scans, and human review beyond basic OCR.
Gaussian Mixture Models in Early Biometric Recognition
How Gaussian mixture models shaped speaker recognition and early biometric research, from GMM-UBM systems to thresholds, errors, and later i-vectors.