2017年

本研究提出一可用於機器人空間探勘與辨識的仿生機器學習系統。本系統模仿生物大腦的層級性結構，各層級間透過雙向連結進行搜尋辨識與提示，並記憶空間中的感官、場景和位置資訊，分別由以下部分構成： 1. 感官細胞：辨識特定感官輸入類別。 2. 場景基模細胞：組合具方向性的感官細胞數據。 3. 網格細胞：接收移動數據的內在座標系統。 4. 位置細胞：整合感官數據與空間數據、建立拓樸空間認知地圖。 經模擬實驗證明，本系統能在第一次探勘時建立空間認知地圖，並於再次造訪時成功匹配位置細胞進行定位。本系統有異地探勘、在複雜空間中進行路徑與任務規劃等廣泛應用。

研究指出葉酸營養狀態與乳癌的風險有相互關聯且癌幹細胞是扮演癌症轉移的主要角色。葉酸缺乏和葉酸補充皆能促進癌細胞的轉移。本研究探討葉酸營養不良和葉酸補充是否會促進乳腺癌細胞轉變成癌幹細胞，將乳癌細胞株養於正常葉酸、低葉酸和高葉酸中，藉以模仿癌症病人的葉酸營養狀態。細胞在每個葉酸濃度中培養2、 4 、 8 天後再培養於腫瘤球體培養液中並測量四種幹細胞標記-mTOR，SHH，Sox2和Nanog。結果顯示:低葉酸會導致癌細胞的生長停滯，藉由標記的表達增加顯示長期低葉酸會促進細胞轉變成癌幹細胞。於腫瘤球體培養液時，低葉酸組亦較對照組於體外產生較大及較多的腫瘤球。高葉酸組會加速癌細胞的增殖和向癌幹細胞的轉變且細胞增生較快並表達較多的標記，其腫瘤球體亦是最大和豐富的。總之，低葉酸和高葉酸都增強了乳腺癌細胞的腫瘤生成和轉變成癌幹細胞。

Tradescantia (Tradescantia fluminensis) is the worst weed in New Zealand. By smothering and shading out seedlings, Tradescantia prevents forest regeneration. Current control methods are ineffective and simultaneously cause harm to native forest. In 2011 Neolema ogloblini, a Brazilian beetle was introduced into New Zealand as a biological control for Tradescantia. To be successful in New Zealand, a country with different environmental factors, the beetles’ ranges of preference (temperature and light intensity) had to be investigated. A gender specific trait also identified, to enable desired sex ratios within founding populations to be selected. [18] This would ensure that the beetles are not released in areas of physiological stress, and can be optimised to have the greatest impact on Tradscantia. To establish how the intensity of light affects the distribution and amount of Tradescantia eaten by N.ogloblini a choice chamber investigation was conducted. Different layers of shade cloth provided a range of light intensities 150-3450Lux (likely to be found under forest canopy where Tradescantia is problematic). Thirty beetles of a range of sizes and approximately same maturity were randomly distributed through the chambers. Each chamber contained a shoot of Tradescantia with 5 leaves. After a 24hour period the number of beetles in each chamber were counted and the amount of surface area of the leaves eaten measured. The effect of temperature on the amount of leaf surface area eaten was investigated by selecting 90 beetles of a range of sizes and withholding food for 24hours. Five beetles were placed in each of three containers containing two leaves. Each trial container was precooled/warmed to the test temperature before the beetles were added. Leaves of a similar size, shape, mass and maturity were used. All leaves were genetically identical and collected from the same location. Sets of three containers were held in the dark at the following temperatures for 24hours: 9°C, 15°C, 20°C, 25°C, 30°C and 35°C. The surface area of leaf eaten at each temperature (mm2) was calculated. Lastly, microscopic dissections were conducted, using 32 beetles ranging in size, to establish if length (measured from the top of the head to the base of the abdomen) could be used as a phenotypic marker to identify beetle gender. While only a very weak positive relationship between increasing light intensity and the number of beetles was found a significantly higher area of leaf was eaten at a light intensity of 3450Lux compared to 150Lux. The amount of leaf area eaten is significantly reduced at temperatures of 15˚C and below, and significantly increased at 35˚C. There is no significant difference in the amount of leaf area eaten when comparing temperatures between 20-30˚C. Females have on average a larger body length (median=4.92mm) than the males (median=4.215mm). Therefore, sites with warmer temperatures in dappled light conditions (3450Lux) should be prioritised for the release of N.ogloblini, as this is the location in New Zealand at which their use as a biological control will be optimised. Beetle length can be confidently used to select desired gender ratios.

Diarrhoea is an infectious disease that claims many lives (particularly among infants) and is known among many other infectious diseases, to have a relationship with climate. This projects quantifies this relationship by comparing health insurance treatment claims for diarrhoea and anti-diarrheal drug sales data from the private health sector with temperature and rainfall data provided by the South African Weather Service. The data were then analysed and show that incidence is highly seasonal and varies among regions with different seasonal climates. The results show that climate variability can explain 73% of diarrhea incidence variation with rainfall explaining 8% and minimum temperature explain the other 65 %. Preliminary predictions of future monthly percentage increases in incidence were then made for a series of time ranges. This project demonstrates that with predicted climate data one can predict future variations of diarrhoea allowing the health sector to be adequately prepared.

由工業4.0智慧生產的啟發，建構機械手臂進行人工智慧黑白棋之對弈系統，透過視訊分析棋面資料，輸入人工智慧黑白棋遊戲判斷。依據結果指揮機械手臂進行落子、取子的動作，人機對奕過程亦會判斷有無不合理的地方，以維持棋奕的規則公平。擷取雲端攝影機的盤面影像，使用霍夫找圓演算法取得棋目位置，透過彩度與明度分析黑子、白子或無子，黑白棋AI程式再透過遊戲樹演算決定落子、取子位置。透過畢氏定理及餘弦定理將棋子位置的立體座標轉成工具座標，再傳送至主控伺服器以指揮機械手臂進行正確的動作。透過減輕重量及使用彈力平衡力矩改善，機械手臂可改善硬體準確度，重現率測試達85%以上。黑白棋AI程式棋力可以與黑白棋app的3級力敵。視訊判斷棋局在調整適當彩度明度後可達100%正確率。透過演算法指揮機器手臂下棋，棋局中完成正確動作可達80%以上的成功率。

雌激素為一種固醇類的環境賀爾蒙，若水生動物長時間暴露於低濃度的雌激素中，即會導致其生理及行為異常。我們企圖從迪化汙水處理廠中分離純化出能有效降解雌激素之菌株，希冀了解這些菌株的生理特性與降解雌激素的能力。首先在活性污泥中加入高濃度雌二醇(1mM)，以增進雌激素降解菌在菌群中的比例，接著利用十倍序列稀釋進行菌株純化。培養期間以薄層層析檢測雌激素降解活性。隨後以固態培養基培養法取得單一菌落，並進行聚合酶連鎖反應及核酸定序。之後我們亦將純化出之菌株進行其他固醇降解測定。目前純化出一株Novosphingobium屬的變形菌，其為一新的菌種，確實能降解雌二醇及其他固醇，包含雌酮、雄烯二酮、膽酸、孕酮和睪酮。然而，其並不能降解乙炔雌二醇。該菌株具有修復受汙染環境及製備雌激素相關藥品之應用潛力。

浮石是一種多孔的火山碎屑岩，氣孔佔總體積很高的比例，因此能浮在水上。本研究針對臺灣新北市金山區神祕海岸所出露沉積層中的浮石進行探討，藉由閱讀文獻、實地考察、採樣、實驗分析等方式，來瞭解浮石的化學組成、礦物種類及可能形成機制與搬運途徑。依照化學成份分析及礦物種類比較結果顯示神祕海岸的浮石與日本、菲律賓的浮石較無關聯，最可能是來自台灣地區並由台灣北部大屯火山系或觀音山噴發所造成。至於其搬運方式，沉積物粒徑大小分析的結果顯示浮石是經由力量強大的海洋搬運作用攜至當地沉積。綜合上述，我們認為金山神祕海岸的浮石最有可能的生成方式為：1867年，由與大屯火山有相同岩漿系統的海底火山噴發出，接著此火山爆發所引發的海嘯將浮石帶至神祕海岸進行沉積。

以製作適合高中生的英文篇章難易度自動分級為初衷，本研究採高中英文課文為語料，針對「如何分級」，意即從文章萃取哪些特徵、利用何工具或語料協助萃取特徵、以何工具分級等因素，進行研究與實驗，並建立一套新方法。首先進行前處理，再嘗試以單字、句型的數量或比例、句長、音節長、整合以上分析等各式特徵，支持向量機(Support Vector Machines)、隨機森林分類器(Random Forest Classifier)、決策樹分類器(Decision Tree Classifier)、卷積神經網路句分類器(Convolutional Neural Networks for Sentence Classification)等工具，進行將篇章分為高中一、二、三年級等三個難易度等級的測試，建立自動分級模型。最後製作成可供大眾使用的自動分級網頁。各項測試之中，最佳分類效能為整合各項特徵時得到的分類正確率65.04%，經模擬得知，此效能較過去研究，已有所提升。

肝癌是死亡率極高的癌症。目前並無非侵入性方法判斷標靶藥物治療效果。蕾莎瓦(Sorafenib)是一種使用於肝癌晚期治療的標靶藥物，但有些病患在治療過程中會出現抗藥性，因此可預測蕾莎瓦治療效果的生物標記相當重要。 血液中的生物標記可及時追蹤疾病進展與預後之評估。本實驗利用質譜法鑑定分析在肝癌細胞(HuH7)中加入蕾莎瓦模擬治療後肝癌的分子機轉，並透過生物資訊軟體了解蛋白質在HuH7細胞中所扮演的角色。我們發現Galectin-3 和HMGB1可能參與腫瘤細胞的增生和遷移，以西方墨點法觀察其表現量，發現Galectin-3 和HMGB1在受到藥物處理後，表現量皆有下降的情形。 蕾莎瓦可抑制B-RAF激酶進而阻斷MAPK pathway。MAPK pathway又可分成ERKs、JNKs和p38/SAPKs三類。為確認Galectin-3之訊息傳遞路徑，將蕾莎瓦和ERK抑制劑處理細胞，發現Galectin-3的表現量隨著時間而下降，證實Galectin-3的表現會受到ERK路徑調控。 Galectin-3與 HMGB1是極具潛力的生物標記可應用於蕾莎瓦的治療，希望這些生物標記可以應用在臨床上。

銦之化合物常被應用於半導體產業當中，隨著使用量的增加也提高銦污染流佈於環境的風險，若此污染物流布至土壤並累積於植物體內，則可能通過食物鏈而影響人體的健康。現今國內外鮮少有銦對植物影響的相關研究，因此本研究以培養基試驗，探討銦對模式植物阿拉伯芥生長之影響。結果顯示，當銦處理濃度為500 µM時，其生質量以及根長會受到抑制。MDA含量的增加，顯示銦會對阿拉伯芥造成氧化逆境，而穀胱甘肽與金屬螯合素並非阿拉伯芥在銦逆境下主要的抗氧化機制。本研究亦發現銦處理下會影響對錳、鉀、鈉、磷、鐵、錳、銅以及鋅的吸收。此外，X光吸收光譜分析顯示根部以及地上部中銦物種有明顯組成上的差異。於根部有42.9%的銦是以吸附於細胞壁的形式存在，這可視為植物對重金屬所採取「隔離」策略的防禦方式之一。