This paper demonstrates the framework and results from the team “Florida International University - University of Miami (FIU-UM)” in TRECVID 2016 [1] Ad-hoc Video Search (AVS) task [2]. The following two runs were submitted: • M D FIU UM.16 1: CNN features + linear SVM + concept scores combination type I • M D FIU UM.16 2: CNN features + linear SVM + concept scores combination type II In both runs, the features are first extracted by the CNN (Convolutional Neural Network) structure of AlexNet [3]. Then, using the linear SVM (Support Vector Machine) classifiers, the scores of each concept for the key frames are generated. For run 1 and run 2, the scores from the aforementioned model are combined in different ways for different queries. From the submission results, run 2 outperforms run 1. The submission details are listed as follows. • Class: M (Manually-assisted runs) • Training type: D (IACC & non-IACC non-TRECVID data) • Team ID: FIU-UM (Florida International University - University of Miami) • Year: 2016
