I have developed an "Inference-by-Composition" approach, which gives rise to analysis and prediction of complex visual information (both in images and in video data) without resorting to any pre-defined parametric models, nor requiring an exhaustive set of prior visual examples (which other approached typically rely upon). Visual "pieces of evidence" from a small number of available visual examples are composed and integrated into new global visual configurations that were never seen before. This allows to make inferences about the likelihood of a combinatorially large set of complex scenes and events that were never observed before. This "Inference by Composition" approach opens the door to analysis and prediction of very complex static and dynamic information, which could not have been previously handled. Moreover, the applicability of this approach extends beyond the field of Computer Vision to multiple disciplines and research areas. More details can be found in my papers [ICCV’05, NIPS’06, IJCV’07, ECCV’08, ECCV’12, ICCV’13, BMVC’14, PAMI’14] and in their corresponding demo webpages.