Adversarial Attacks Against Detecting Bot Generated Text
With the introduction of the transformer architecture by Vaswani et al. (2017), contemporary Text Generation Models (TGMs) have shown incredible capabilities in generating neural text that, for humans, is nearly indistinguishable from human text (Radford et al., 2019; Zellers et al., 2019; Keskar et al., 2019). Although TGMs have many potential positive uses in writing, entertainment and software development (Solaiman et al., 2019), there is also a significant threat of these models being misused by malicious actors to generate fake news (Uchendu et al., 2020; Zellers et al., 2019), fake product reviews (Adelani et al., 2020), or extremist content (McGuffie & Newhouse, 2020). TGMs like GPT-2 generate text based on a given prompt, which limits the degree of control over the topic and sentiment of the neural text (Radford et al., 2019). However, other TGMs like GROVER and CTRL allow for greater control of the content and style of generated text, which increases its potential for misuse by malicious actors (Zellers et al., 2019; Keskar et al., 2019). Additionally, many state-of-the-art pre-trained TGMs are available freely online and can be deployed by low-skilled individuals with minimal resources (Solaiman et al., 2019). There is therefore an immediate and substantial need to develop methods that can detect misuse of TGMs on vulnerable platforms like social media or e-commerce websites. Several methods have been explored in detecting neural text. Gehrmann et al. (2019) developed the GLTR tool which highlights distributional differences in GPT-2 generated text and human text, and assists humans in identifying a piece of neural text. The other approach is to formulate the problem as a classification task to distinguish between neural text and human text and train a classifier model (henceforth a ‘detector’). Simple linear classifiers on TF-IDF vectors or topology of attention maps have also achieved moderate performance (Solaiman et al., 2019; Kushnareva et al., 2021). Zellers et al. (2019) propose a detector of GROVER generated text based on a linear classifier on top of the GROVER model and argue that the best TGMs are also the best detectors. However, later results by Uchendu et al. (2020) and Solaiman et al. (2019) show that this claim does not hold true for all TGMs. Consistent through most research thus far is that fine-tuning the BERT or RoBERTa language model for the detection task achieves state-of-the-art performance (Radford et al., 2019; Uchendu et al., 2020; Adelani et al., 2020; Fagni et al., 2021). I will therefore be focussing on attacks against a fine-tuned RoBERTa model. Although extensive research has been conducted on detecting generated text, there is a significant lack of research in adversarial attacks against such detectors (Jawahar et al., 2020). However, the present research that does exist preliminarily suggests that neural text detectors are not robust, meaning that the output can change drastically even for small changes in the text input and thus that these detectors are vulnerable to adversarial attacks (Wolff, 2020). In this paper, I extend on Wolff’s (2020) work on adversarial attacks on neural text detectors by proposing a series of attacks designed to counter detectors as well as an algorithm to optimally select for these attacks without compromising on the fluency of generated text. I do this with reference to a fine-tuned RoBERTa detector and on two datasets: (1) the GPT-2 WebText dataset (Radford et al., 2019) and (2) the Tweepfake dataset (Fagni et al., 2021). Additionally, I experiment with possible defences against these attacks, including (1) using count-based features, (2) stylometric features and (3) adversarial training.
Optimization of honey production by monitoring the behavior of bees based on studying their sounds
This is a first approach in the development of beekeeping and the preserving of bees, a crucial and important species in the balance of ecology on our planet. This project consists in designing and building a small affordable device that will help beekeepers keep an eye on their hives and prevent theft whenever and wherever they are by providing them with instant and continuous data and information about their beehive status through a mobile application. This IOT approach will rely on many physical variables especially the sound frequency of the bee buzz, which appears to be a way for the bees to communicate with each other in special circumstances. That is why; we aimed to analyze the sound frequencies of the bee buzz to detect beehive behavioral changes. Many other factors are also important for the keeping of a healthy beehive such us temperature, humidity, weight and fly activity. And as for security measures we are going to add a GPS tracker to the system to keep track of the hives and alert the beekeeper if there is any kind of danger. The development of this real time beehive monitoring system will not only help the beekeeper keep track of his hive and collect useful data but also increase the honey production and avoid many colony losses and thus preserve the bees and ensure their well-being.
隱密的發育調節中樞-植物轉錄因子BPC對發育之調控機制 A cryptic hub for development control: Unraveling the regulatory role of plant transcription factor class I BASIC PENTACYSTEINEs in Arabidopsis development
GAGA 序列為生物發育重要順式作用子； BPC (BASIC PENTACYSTEINE) 則為植物特有 GAGA 結合蛋白。已知 bpc 突變體具多效性，其生理時鐘相關之發育有多重缺陷。阿拉伯芥BPC家族中 BPC1, BPC2, BPC3 為第一亞群，且 BPC 群間和群內有重疊與拮抗作用。為探究第一群 BPC 是否調控生理時鐘，本實驗以 3D 影像觀察 bpc1 bpc2、bpc1 bpc2 bpc3 及野生型之晝夜運動，並誘導 BPC 過量表現以檢測時鐘基因反應，發現 bpc 突變體之晝夜運動與時鐘節律皆有缺陷，顯示 BPC 能影響生理時鐘運行。透過一系列對第一群 BPC 突變體與過量表現植株的 RT-qPCR 檢測，可歸結第一群 BPC 是能調控生理時鐘與葉片生長的中心。
DEVELOPMENT OF PAPER-BASED ORIGAMI BIOSENSOR PLATFORMS FOR COLORIMETRIC DETECTION OF BIOCONTAMINANTS
Infectious diseases caused by bacteria from biological pollutants pose a great burden in terms of diagnosis and treatment, and millions of people worldwide die from bacterial infections. Detection of bacteria plays a critical role in clinical diagnosis and control of contamination, but is not accessible due to the high cost, complex devices and equipment required. In the project, an alternative to existing methods, a paper-based biosensor for the detection of model organism E. coli bacteria, which is visible, low cost, easy to use, can be integrated with a smartphone, is based on rapid color change in the exposed environments, drinking and pool water, wastewater, beverage products. platforms were developed. For the specific detection of E.coli bacteria, two different biosensors have been developed that can perform colorimetric detection in a user-friendly origami design, minimizing microchip and processing steps based on antibody-bound PVDF membrane and filter paper-based immunological method. In the presence and absence of target bacteria E.coli, the lowest detection limit of the biosensors obtained by using paper-based platforms that create a distinctive color on them, depending on the concentration, was 0.9x103 bacteria/ml for origami biosensor, 2.7x103 bacteria/ml for microchip biosensor and the widest dynamic linear operating range was calculated as 103-107 bacteria/ml. With the biosensor platforms we have developed, the use of only one smartphone for both qualitative and quantitative, visible results and analysis within minutes constitutes the originality of our project. With these promising results, the biosensors we have developed can also be used for the detection of different biological pollutants, do not contain complex devices and can be easily produced in large scales. We believe that the biosensors we have developed for the detection of biological pollutants in water and beverages, especially in regions where test laboratory infrastructure is not available, will contribute to the literature, public health, health economy and sustainable development goals such as clean water and sanitation, health and quality life, and life in water.
Synthesize Sodium Sesquicarbonate and Increase Yield
In order to recycle disposable diapers, we investigated the conditions where sodium sesquicarbonate (Chemical formula Na2CO3・ NaHCO3・ 2H2O hereinafter called sesqui) precipitates selectively from sodium carbonate and the conditions for high yield. For the selective precipitation of sesqui, we defined the time required for the reaction solution to pass through the sesqui precipitation area in the Na2CO3-NaHCO3-H2O phase diagram (45°C) as Δ t. As a result, we revealed that Δt is involved in the selective precipitation of sesqui, and that we can synthesize sesqui without the expensive addition of L-Arginine as used in a previous research. Also, we proposed the “Stay method”, in which the supply of CO2 is stopped for 30 minutes to the lengthen the Δ t, and found that we could synthesize sesqui selectively even under conditions in which sodium bicarbonate is likely to be precipitated as well. Regarding the high yield of sesqui, the yield was greatly improved by the common ion effect of Na by adding NaOH to the reaction solution, sesqui synthesis by repeated reactions with CO2, and sesqui recovery by adding the anti-solvent ethanol, reaching a sesqui conversion rate of 95%. This means 109 g of sesqui can be synthesized from 100 g of Na2CO3. Moreover, we confirmed that these synthesized samples have almost the same detergency as commercial sesqui. We did a test calculation to reveal the usefulness of this research. First, if diaper recycling technology is put into practical use and all used diaper waste in Saijo City can be recycled, a reduction of 534 t/year of used diaper waste can be expected. This corresponds to a 2.3% reduction in Saijo City's waste output. From the ash that would ultimately remain after being recycled, we expect up to 35.3 t/year of synthesized sesqui using our experimental method. In addition, a CO2 reduction of 8.2 t/year is possible in the process, which is about equivalent to the volume of one gymnasium.