Stephen Scott

About

Stephen Scott is a leading researcher in computational learning theory, with a primary focus on multiple-instance learning (MIL) and PAC (Probably Approximately Correct) learning. His most influential contribution is the 2005 paper “On Generalized Multiple-Instance Learning,” which has garnered 68 citations. In this work, Scott introduced a novel generalization of the traditional MIL framework, where a bag is labeled positive only if it contains a collection of instances, each near one of a set of target points—a significant departure from the standard single-target model. This contribution has broadened the theoretical foundations of MIL, impacting applications in drug discovery, image classification, and object detection. Scott’s earlier work on noise-tolerant algorithms for learning geometric patterns (1999, 14 citations) and PAC learning of one-dimensional patterns (1996, 9 citations) further demonstrates his sustained commitment to understanding how machines can learn from imperfect data. His research is characterized by rigorous theoretical analysis paired with empirical validation, making his findings both principled and practical. For students and researchers, Scott’s work offers a clear window into the evolution of learning theory and the power of rethinking core assumptions in machine learning.

Research Focus

Key Achievements

Top Papers

1. 1
   ON GENERALIZED MULTIPLE-INSTANCE LEARNING

   68 citations · 2005
2. 2
   A Theoretical and Empirical Study of a Noise-Tolerant Algorithm to Learn Geometric Patterns

   14 citations · 1999
3. 3
   PAC Learning of One-Dimensional Patterns

   9 citations · 1996
4. 4
   PAC learning of one-dimensional patterns

   3 citations · 1996

Key Collaborators

Contact & Links