Abstract. We propose a novel supervised learning technique for summarizing videos by automatically selecting keyframes or key subshots. Casting the task as a structured prediction problem, our main idea is to use Long Short-Term Memory (LSTM) to model the variable-range temporal dependency among video frames, so as to derive both representative and compact video summaries. The proposed model successfully accounts for the sequential structure crucial to generating meaningful video summaries, leading to state-of-the-art results on two benchmark datasets. In addition to advances in modeling techniques, we introduce a strategy to address the need for a large amount of annotated data for training complex learning approaches to summarization. There, our main idea is to exploit auxiliary annotated video summarization datasets, in spite of their heterogeneity in visual styles and contents. Specifically, we show that domain adaptation techniques can improve learning by reducing the discrepancies in the original datasets' statistical properties.

No abstract

Most studies on visual working memory (VWM) and spatial working memory (SWM) have employed visual stimuli presented at the fronto-parallel plane and few have involved depth perception. VWM is often considered as a memory buffer for temporarily holding and manipulating visual information that relates to visual features of an object, and SWM for holding and manipulating spatial information that concerns the spatial location of an object. Although previous research has investigated the effect of stereoscopic depth on VWM, the question of how depth positions are stored in working memory has not been systematically investigated, leaving gaps in the existing literature on working memory. Here, we explore working memory for depth by using a change detection task. The memory items were presented at various stereoscopic depth planes perpendicular to the line of sight, with one item per depth plane. Participants were asked to make judgments on whether the depth position of the target (one of the memory items) had changed. The results showed a conservative response bias that observers tended to make 'no change' responses when detecting changes in depth. In addition, we found that similar to VWM, the change detection accuracy degraded with the number of memory items presented, but the accuracy was much lower than that reported for VWM, suggesting that the storage for depth information is severely limited and less precise than that for visual information. The detection sensitivity was higher for the nearest and farthest depths and was better when the probe was presented along with the other items originally in the memory array, indicating that how well the to-be-stored depth can be stored in working memory depends on its relation with the other depth positions.

Post-operative cognitive dysfunction (POCD) could cause short-term or long-term cognitive disruption lasting weeks or months after anesthesia and surgery in elderly. However, no effective treatment of POCD is currently available. Previous studies indicated that the enhancement of brain-derived neurotrophic factor (BDNF) expression, and the elevation the cholinergic system, might be effective to prevent POCD. In this study, we have discovered that tacrine(10)-hupyridone (A10E), a novel acetylcholinesterase (AChE) inhibitor derived from tacrine and huperzine A, could prevent surgery-induced short-term and long-term impairments of recognition and spatial cognition, as evidenced by the novel object recognition test and Morris water maze (MWM) tests, in aged mice. Moreover, A10E significantly increased the expression of BDNF and activated the downstream Akt and extracellular regulated kinase (ERK) signaling in the surgery-treated mice. Furthermore, A10E substantially enhanced choline acetyltransferase (ChAT)-positive area and decreased AChE activity, in the hippocampus regions of surgery-treated mice, indicating that A10E could prevent surgery-induced dysfunction of cholinergic system, possibly via increasing the synthesis of acetylcholine and the inhibition of AChE. In conclusion, our results suggested that A10E might prevent POCD via the activation of BDNF pathway and the inhibition of AChE, concurrently, in aged mice. These findings also provided a support that A10E might be developed as a potential drug lead for POCD.

The aim of this study was to identify and elucidate the vasorelaxant activity of echinacoside, a phenylethanoid glycoside isolated from the medicinal herb Cistanche tubulosa, and its possible underlying mechanism on isolated rat thoracic aortic rings pre-contracted with phenylephrine (PE, 1 microM) and KCl (60 mM). Echinacoside (30-300 microM) exhibited an acute relaxation in endothelium-intact rings in a concentration-dependent manner, while this relaxation was significantly inhibited in endothelium-denuded condition and in the presence of the endothelial nitric oxide synthase (eNOS) inhibitor, N(W)-nitro-L-arginine methyl ester (L-NNA, 100 microM), an unselective soluble guanylate cyclase blocker, methylene blue (10 microM), the selective sGC inhibitor 1 H-[1, 2, 4]oxadiazolo[4,3- A]quinoxalin-1-one (ODQ, 1 microM); in addition, atropine (1 microM), a selective muscarinic receptor antagonist, partially affected the relaxation. However, the cyclooxygenase inhibitor indomethacin (5 microM) had no influence on the action. Echinacoside enhanced the cyclic guanosine monophosphate (cGMP) production in aortic rings contracted with PE. These results indicate for the first time that echinacoside mediates the endothelium-dependent vasodilator action in rat thoracic aortic rings through nitric oxide (NO)-cGMP pathway.