2018年

全球性氣候變遷，一般的種植常因無法順應調節而產生病害，合併蟲害傳染時更須使用化學農藥。現今新興的農業設施有溫室栽培、魚菜共生及植物工廠等，但仍存在著介質控管、溫濕度控制及成本過高等問題。 本研究自行設計製作「植物箱」，運用替代光源及冷熱蒸散原理，模擬取代自然環境中之最佳光照、溫度與溼度，建立適宜植物生長的環境。實驗結果顯示：一、T5燈管能模擬陽光的輻射熱能和光度，配合紅藍光比例及定時器，可控制光照時間及取代陽光照射。二、水冷機控制貯水槽溫度，結合風扇向上吹送，有助水氣蒸發與凝結，達到溫度控制。三、溼度控制器結合風扇向下吹送，有助水氣吹送，形成循環對流，達到溼度控制。四、自製密合及低功率的「植物箱」為智能環控，可避免病蟲害及農藥的使用。 在未來可創造極大的經濟價值，以更節能環保的方式與地球永續共存。

我們製作獨特的垂直風洞，透過牙條及滾珠軸承使羽球能在其中自由轉動以利同步取得運動時的阻力、轉速及側向力，並透過空氣壓縮噴射方式探討軌跡差異。實驗數據顯示，羽毛只要經過修剪，阻力多會下降，雖然能飛的更高遠，但也降低了羽球獨有的飛行特性。若修剪外側羽毛，可大幅提升高速下的轉速，但低速時則因受力面積減少導致轉速下降且發現透過側羽的小修剪可同時提升飛行能力以及飛行特性，提升練習級羽球的價值；但羽尾若破壞程度太大，如：剪成平頭狀，飛行距離雖明顯提升，但轉動能力下降，甚至低速時無法轉動。最後，斷翅的實驗出現了一個有趣的結果：透過斷去第二根羽毛達到對稱的缺口，可挽救失控的側向力，提升破損羽球的練習價值。

本次研究中，我使用Arduino 單晶片微控制板自製了細懸浮微粒偵測器，用來測量PM2.5的穿透率，以達量化懸浮微粒的效果，實驗中發現，使用摩擦起電的方式因為電場太小無法有效吸附懸浮微粒，接著使用平行電網，發現吸附能力與電壓與總表面積有關，因而改採電纖維通以高壓電的方式進行濾淨，其吸附率約達52%。在整個實驗中我使用Arduino單晶片微控制板控制數據的擷取、分析，與節電系統的調控，做出了一個兼具輕便、低成本、與節電環保的懸浮微粒清淨器。

大數據時代促成運動賽事分析更加蓬勃發展，一般來說數據分析需要三個要素：資料、分析演算法和應用領域知識；隨著開放資料分享的普遍化，人們可以更輕易的取得資料並運用各式分析演算方法來針對有興趣的應用問題做出更精準的預測。 本研究以kaggle平台上所提供之歐洲職業足球比賽之公開資料集為基礎，使用目前最具分析潛力的深度學習技術─結合卷積神經網路（CNN）和全連接型神經網路的設計出一個五層的學習架構，建置出足球比賽結果的預測模型。此模型可直接預測主隊勝、負以及平手等三種結果，實驗結果亦展示出本研究所建置的SoccerNet預測模型優於過往的研究，有著更佳的預測能力；同時也驗證了使用公開資料集與CNN技術在球賽分析的可能性。而本研究所提的SoccerNet模型不僅可以運用於賽前的結果預測，亦可運用於球隊經營管理等決策，頗具有商業價值。

An increase in blood cholesterol contributes to cardiovascular diseases, the number one cause of death worldwide. Statins are currently the most effective in reducing cholesterol levels and treating patients with high cholesterol. However, these pharmaceutical agents have been shown to cause several side effects, prompting the need for a more natural solution to increasing cholesterol levels. Hence, a study was conducted to investigate the ability of lactic acid bacteria in the removal of cholesterol, explore the mechanism for the removal of cholesterol by lactic acid bacteria, and examine the effectiveness of kidney beans and sunflower seeds in inhibiting HMG-CoA reductase in the cholesterol biosynthesis pathway. Results showed that Lactobacillus plantarum was the most effective in reducing cholesterol levels and that the mechanism for cholesterol removal included both the binding to cell wall and active uptake into cells. Sunflower seeds and kidney beans were also shown to be effective in inhibiting HMG-CoA reductase, with sunflower seeds having 100% inhibition of the enzyme, similar to pravastatin, a commercial cholesterol reducing drug, and kidney beans having comparable percentage inhibition of the enzyme compared to pravastatin.

The purpose of this research is to make an ecological insecticide that mixes the extracts of Piper tuberculatum, Annona muricata and Melia azedarach, that together in application cause mortality and repellence of the mosquito Aedes aegypti with the intention to help in the control of the diseases this mosquito is guilty of: dengue, zika and chikungunya, and decrease the risk of infection by a safe and organic way.

This study demonstrates that microbial film power generation is a potentially viable source of alternative energy. This research occurred over a period of two years. In the first year (2016) I tested a new method of generating renewable energy, referred to as microbial film power generation. I showed that electricity could be captured from microbial decomposition using solid graphite plates (29cm x 20cm) placed in lightly decomposed muskeg (collected in northern British Columbia). In the second year (2017) the purpose was to increase the power output of the fuel cell, while also compacting the setup. Certain changes were made to the experimental set up, namely the use of spongy graphite felt in place of solid graphite plates, thus providing a larger surface area for microbial activity to occur. The new fuel cells made produced about twice as much power. Not only was the power output greater, but it was produced from a much smaller area: 7.82 mWh/cm2 on graphite felt, compared to 0.21 mWh/cm2 on graphite plates. In other words, graphite felt produced 37 times more power per unit area than graphite plates. Furthermore, it would appear that by removing the load from the fuel cell for approximately 24 hours, the fuel cell could essentially recharge. This may be due to microbial activity releasing more electrons onto the anode permitting a new cycle to take place. This would suggest that the system could naturally recharge itself.

The rate and production of conventional petroleum based plastics is unsustainable and not eco-friendly. Plastics often end up in marine environments and can take hundreds of years to decompose in landfills. According to Statistica, in 2015 alone, global plastic production was approximately 322 million metric tonnes and is projected to increase in the future. PHB bioplastic or Polyhydroxybutyrate is both biologically produced and biodegradable and can serve as a viable alternative to conventional plastics. But can it be broken down by soil microbes within a reasonable time frame? I have set out to answer this question. My aim was to isolate and analyse microorganisms from the Rotorua area that are capable of degrading Polyhydroxybutyrate (PHB) bioplastic . I isolated PHB degrading microorganisms from Rotorua soils by culturing on an agar based mineral salt media supplemented with PHB powder (MSM PHB agar). Samples were taken from Mount Ngongotaha and Te Puia geothermal soils as well as Okareka, termite frass and termite guts. One isolate from the Te Puia sample (labelled G2) was found to successfully degrade PHB powder. After isolation and purification of the G2 isolate, it was cultured on a range of media types to examine properties exhibited under differing nutrient conditions. Multiple organisms were found to be involved in the degradation of PHB bioplastic and work together symbiotically, this included bacteria and fungi which was identified as penicillium. The sample isolated from Te Puia soils (site 2 – G2Clear) in the Rotorua environment was found capable of competently degrading PHB, clearing 8% of PHB after 26 days. The G2Clear isolate is a mixture of bacteria and fungi working in an endosymbiotic relationship to degrade PHB and are unable to successfully degrade PHB individually. It is through the secretion of an extracellular PHB depolymerase enzyme that PHB is degraded, conforming with my hypothesis. This proves that PHB bioplastic is a viable alternative to conventional petroleum based plastics as PHB can be relatively quickly broken down by soil microorganisms.

深度學習透過迭代的訓練，如沿著梯度反方向更新權重的梯度下降法，旨在找出損失函數建構出的多維函數圖形中，其全局最小解的變數組合。我們受到在睡眠中的快速動眼期啟發，此時期腦中的高頻率γ波可以強化學習效果。本研究模擬此現象，提出一使原損失函數擁有自適應的增幅功能的演算法（稱之為REM方法），且其中的超參數能夠根據其應用調整。 本研究將REM方法應用於三種經典的優化法，並且以五種異質的資料集測試。實驗結果指出，在搭配隨機梯度下降法（SGD）與自適應學習率優化法（Adagrad）時，REM有顯著的優化效果。REM方法不僅能大幅加速訓練進程，亦能避免特定的訓練問題。

本研究利用回饋機制修剪人工神經元的方式，可在較短時間內達到較高的目標行為學習正確次數。這樣的系統設計源於生物行為的啟發，因為回饋機制能直接影響決策中樞的神經元，而非只是影響動作本身， 在達成目標學習上會更直接、快速。以傳統強化學習 （Reinforcement Learning）系統為例，本研究模擬結果顯示，若回饋機制影響動作本身，目標達成正確率只有 19.14％。而若獎勵回饋到神經元的修剪上，則相較強化學習提高超過 2 倍的正確率。甚至，隨機修剪神經元也可相較強化學習提高 1 倍的正確率。顯然地，本系統能確實提高正確次數並縮短目標達成時程。利用回饋機制修剪人工神經元，可為強化學習在目標學習上遇到的困境，提供一個新的思考方向，實務應用上，可彌補強化學習在學習行為上無法一般化的缺點。