Stacking-based deep neural networks for pattern classification

Low, Cheng-yaw Graduate School, Yonsei University 2018 국내박사

이 논문은 스택 기반의 심층 신경망(Stacking-Based Deep Neural Networks, S-DNN)을 심층 신경망들의 한 종류로 정의한다. 고정된 아키텍쳐와 단대단의 적연적 비용함수로 설정되는 기존의 심층 신경망 모델들과 달리, S-DNN은 자체학습이 가능한 여러 모듈들의 조합으로 구성된 신경망이다. 부분망이거나 각산법(interleaving) 방식의 층인 각 모듈은 대규모 문제를 근사법과 역전달 없이 조합으로 풀어낼 수 있다. 이 논문은 S-DNN으로부터 파생된 다음의 세 개의 모델들을 소개한다. (1) 다층 레이어 스택의 플래튼(flattened) 방법으로 생성된 단층 레이어의 다차원 필터간 컨볼루션 망과 (2) 심층 주 성분 분석망과 (3) 리지회귀(ridge regression) 기반의 심층 분석망(Deep Analytic Networks, DAN)이 소개 모델들이다. 처음 두 개의 망들은 컨벌루션 기반이고, 마지막 DAN은 비컨벌루션 형태의 (즉, 완전 연결 형태의) 일반적 모델이다. M-FFC는 필터 문제를 해결하기 위해 개발되었고, PCANet+ 은 비선형성질의 결핍으로 인한 망 플래트닝 문제와 특징 과다로 인한 깊이 한계 문제를 해결하기 위해 설계되었다. 그리고 DAN은 이미 학습된 회귀 결과에 다시 학습함으로써 성능 증가를 도모한다. 우리는2-FFC, PCANet+, DAN모델의 성능평가를 진행하였고 이 성능 평가는 얼굴인식 뿐만아니라 일반적인 객체도(문자, 자동차, 동물 등) 이용하여 우리는 평가되었다. 그 외에도 사전 추출된 특징들과 수정된 특징들에 대한 데이터에도 모델 성능을 측정하였다. 실험결과를 통해, 2-FFC 망과 PCANet+ 을 통해 얻어진 특징이 다른 S-DNN 기반의 망들을 통해 얻어진 특징보다 좋다는 것을 보였다. 또 DAN에 대한 실험에서는, 입력 데이터 종류의 무관하게 레이어가 깊어짐에 따라 성능이 증가하는 것을 관찰할 수 있었다. This dissertation defines stacking-based deep neural networks (S-DNN) as a resemblance to deep neural networks (DNN). Contrary to DNN learning a pre-fixed deep networks in the end-to-end setup with respect to the global loss function, S-DNN is a deep composition of self-learnable modules. Each modular unit for S-DNN, either a subnetwork or a non-linear interleaving layer, deciphers a large-scale problem decisively and independently, with no backpropagation intervention. This dissertation outlines three S-DNN derivatives, including (1) the single-layer multi-fold filter-to-filter convolution (M-FFC) networks as a multi-layer flattened S-DNN instance; (2) the deep principal component analysis networks (PCANet+); and (3) the ridge regression-driven deep analytic networks (DAN) (and also its kernelization, K-DAN). The first twos are of convolutional manipulating only images; and DAN (and K-DAN) is rather non-convolutional (and therefore fully-connected) accepting features of non-specific types. In general, M-FFC is devised to confront with the filter scarcity problem suffering most of the shallow filter bank-based S-DNNs such that an adequate number of filters are learned and diversified for convolutional feature extraction. In the meantime, PCANet+ targets to untangle the network flattening issue (as no non-linearity presents in between two adjacent convolutional layers), and the network depth restriction (due to feature explosion) for the existing S-DNNs. Along with that, by aggregating the layer-wise ridge regression outputs deep, DAN triggers feature re-learning which in turn leads to performance gain. The 2-FFC, the PCANet+, and the DAN performance in pattern classification are scrutinized, not only limited to faces, but also other generic objects. Aside from that, the pre-extracted features (derived based on the established feature encoder), and the human-engineered features are also considered. The empirical results demonstrate that the 2-FFC and the PCANet+ descriptors prevail over other pertinent S-DNNs. Furthermore, DAN (and K-DAN) escalates the layer-wise performance as navigating into the sufficiently deep networks, disregarding of the input feature type.

Automated Performance and Correctness Debugging for Big Data Analytics

Teoh, Jia Shen ProQuest Dissertations & Theses University of Cali 2022 해외박사(DDOD)

The constantly increasing volume of data collected in every aspect of our daily lives has necessitated the development of more powerful and efficient analysis tools. In particular, data-intensive scalable computing (DISC) systems such as Google’s MapReduce [36], Apache Hadoop [4], and Apache Spark [5] have become valuable tools for consuming and analyzing large volumes of data. At the same time, these systems provide valuable programming abstractions and libraries which enable adoption by users from a wide variety of backgrounds such as business analytics and data science. However, the widespread adoption of DISC systems and their underlying complexity have also highlighted a gap between developers’ abilities to write applications and their abilities to understand the behavior of their applications. By merging distributed systems debugging techniques with software engineering ideas, our hypothesis is that we can design accurate yet scalable approaches for debugging and testing of big data analytics’ performance and correctness. To design such approaches, we first investigate how we can combine data provenance with latency propagation techniques in order to debug computation skew —abnormally high computation costs for a small subset of input data —by identifying expensive input records. Next, we investigate how we can extend taint analysis techniques with influence-based provenance for many-to-one dependencies to enhance root cause analysis and improve the precision of identifying fault-inducing input records. Finally, in order to replicate performance problems based on described symptoms, we investigate how we can redesign fuzz testing by targeting individual program components such as user-defined functions for focused, modular fuzzing, defining new guidance metrics for performance symptoms, and adding skew-inspired input mutations and mutation operation selector strategies.For the first hypothesis, we introduce PERFDEBUG, a post-mortem performance debugging tool for computation skew—abnormally high computation costs for a small subset of input data. PERFDEBUG automatically finds input records responsible for such abnormalities in big data applications by reasoning about deviations in performance metrics such as job execution time, garbage collection time, and serialization time. The key to PERFDEBUG’s success is a data provenancebased technique that computes and propagates record-level computation latency to track abnormally expensive records throughout the application pipeline. Finally, the input records that have the largest latency contributions are presented to the user for bug fixing. Our evaluation of PERFDEBUG using in-depth case studies demonstrates that remediation such as removing the single most expensive record or simple code rewrites can achieve up to 16X performance improvement.Second, we present FLOWDEBUG, a fault isolation technique for identifying a highly precise subset of fault-inducing input records. FLOWDEBUG is designed based on key insights using precise control and data flow within user-defined functions as well as a novel notion of influence-based provenance to rank importance between aggregation function inputs. By design, FLOWDEBUG does not require any modification to the framework’s runtime and thus can be applied to existing applications easily. We demonstrate that FLOWDEBUG significantly improves the precision of debugging results by up to five orders-of-magnitude and avoids repetitive re-runs required for post-mortem analysis by a factor of 33 compared to existing state-of-the-art systems.Finally, we discuss PERFGEN, a performance debugging aid which replicates performance symptoms via automated workload generation. PERFGEN effectively generates symptomproducing test inputs by using a phased fuzzing approach that extends traditional fuzz testing to target specific user-defined functions and avoids additional fuzzing complexity from program executions that are unlikely unrelated to the target symptom. To support PERFGEN, we define a suite of guidance metrics and performance skew symptom patterns which are then used to derive skew-oriented mutations for phased fuzzing. We evaluate PERFGEN using four case studies which demonstrate an average speedup of at least 43X speedup compared to traditional fuzzing approaches, while requiring less than 0.004% of fuzzing iterations.

Interpretation of cultural landscape in traditional Malay village : case study of Padang changkat village in Malaysia

Teoh, Meiyee 서울대학교 대학원 2013 국내석사

This thesis is an ethnographic study of describing the cultural landscape of traditional Malay village in Malaysia called Padang Changkat. The study focuses on the characteristics and implication of cultural landscape, in order to know the way that indigenous people build their home settlement, which influence and develop their original perception of home-making, so that the home settlement can express who they are, help people become who they are meant to be, and finally it becomes one of their ethnic expression. This study has three objectives: first, to determine the physical and social attributes that shaped the cultural landscape in Malay village; second, to analyze the specific logic and implication behind the landscape patterns inside the Malay village that contributed to the identity of place; third, to interpret the Malay home settlement landscape in aspects of physical and social-cultural context. To achieve the objectives, this study has been undertaken through several phases which consist of: literature review, research methods, data collecting and analysis of the study site, and finally resulted in the interpretation form. Throughout the study, several methods are adopted to enable the exploration and understanding on the identities of Malay village (kampung) studied. The overall study is underlay on the basis of COSI (causation, operationalization, sampling, and interpretation) method that has been common used in cultural psychology studies. In this study, it is supported by the participant observation as field study method, as well as the hermeneutics approach for the final interpretation phase. To fit the study into the native context, research measurements are decided based on narrative knowledge derived by Malay kampung own vernacular system. They have been identified and categorized as: 1. Background (for macro analysis) 2. Villagers (for macro analysis) 3. Overall kampung layout (for macro analysis) 4. Malay houses (for micro analysis) 5. Malay homegarden (for micro analysis) 6. Orchard farms and forest (for micro analysis) 7. Winding paths and roads (for micro analysis) 8. River (for micro analysis) 9. Historical and cultural elements (for micro analysis) Resulted from interpretative analysis, each feature inside Padang Changkat has demonstrated different faces and dimension of Malay village cultural landscapes; even though these features are united into a whole and difficult to be looked separately in actual state, especially for outsiders. As summary of each feature mentioned above, first, the background of Padang Changkat has an intimate bond with the local royal called Perak Sultanate as well as development of Perak State, which directly related to the presence of village as well as its craftsmanship tradition. Second, villager is the dominator of environment and culture of Padang Changkat through the time-honored coordination among the nature and the society itself. Third, in terms of village layout, Padang Changkat is basically laid within the circle of reserved forest and large-scaled orchard farms that surrounded by a bow–like curve of river; and formed by a centered bulk of Malay houses, which are snuggled among the winding paths, thickets of fruit trees and edible plants simultaneously. Forth, Malay house is the most idiomatic representation of the Malay traditional living culture- it not only represent the native principles in term of “spirit” and “life” of traditional rural Malays, but also constitute a consummate system for both daily activities and social purpose. Fifth, the Malay homegarden as well as orchard farms and forest most reflect the living landscape in Padang Changkat when the multi-layered and complex planting composition not only fulfill the local necessities of life, but also demonstrate the traditional ethno-botany knowledge of rural Malay as well. Sixth, the winding paths and roads provide maximum flow ability and space flexibility; meanwhile they also act as social place for the villagers as well. Seventh, Perak River, which is the river laid at the edge of Padang Changkat, has several significant meanings ranged from politic and economic importance in the old times, to the present-day historical, socio-cultural and recreational value for the village as well as the Perak Royal. Lastly, the historical and cultural elements found in Padang Changkat have functioned as the reinforcing agent for the significance of Padang Changkat cultural landscapes. The overall study of Padang Changkat in fact explicates the Malay village characteristics and implication through its spaces, elements, people and activities. The interpretation has showed how they are functioned in collectively so that formed a vernacular living system in rural Malaysia, as an ethnic responsive landscape specifically for Malays. Finally, this study recommends several schemes for Padang Changkat future development, which considering the rural village improvement as well as cultural heritage conservation and development. The schemes include the use of modern ways in traditional context; declaration as traditional folk village; and application of tradition as authority. These ideas are all underlay on the Frampton’s idea of Critical Regionalism, which mainly focused on the cultivation of resistant and identity-giving culture, with a discreet recourse to modern techniques. As conclusion, both in-depth understanding on Malay village and interpretation of its cultural landscape, in fact, function as ‘value statement’ for rural settlement landscape’s significance to reinforce its survival in the modern era. Therefore, this study, lastly, can be used to solidify the justifiability of calling for attention on the conservation and development of rural village cultural landscape in Malaysia in the future research and planning. 본 연구는 문화기술학적 연구방법 (ethnographic study)을 채택하여 말레이시아의 말레이 전통마을인 파당창카드 (Padang Changkat) 의 문화경관을 해석하였다. 특히 문화경관의 특징과 영향을 중심으로 연구가 진행되었다. 이 지역만의 주민들의 주거 환경은 자신이 누구인지, 또 다른 사람들과 모여서 어떠한 지역 문화를 형성하는지를 보여주기에 파당창카드의 문화경관을 이해하는데 그들의 주거환경에 초점을 두고 연구를 진행하였다. 본 연구는 말레이시아 전통마을의 문화경관을 해석하는데 있어 다음과 같은 목적을 가지고 있다. 첫째, 말레이 마을의 문화경관을 형성하는 물리적, 사회적 환경요인을 파악한다. 둘째, 말레이 마을만의 물리적 경관 패턴을 바탕으로 형태에서 의미를 파악하고 분석한다. 셋째, 물리적, 사회문화적 맥락에서 말레이 주거 정착 마을의 경관을 해석한다. 이와 같은 목적을 가지고 본 연구는 이론적 고찰, 방법론 연구, 대상지 자료 수집 및 분석, 해석하는 과정으로 진행하였다. 이를 위해 본 연구에서는 문화인류학적 연구방법을 비롯한 여러 연구방법으로 말레이 전통마을(kampung)을 연구하고 해석했다. 전반적으로 문화적 심리 연구에서 사용되는 COSI (Causation 원인, Operationalization 시행, Sampling 표본조사, Interpretation 해석) 연구방법을 채택한다. 대상지 현장에서의 연구방법은 Spradley (1980)의 참여관찰법, 말레이 마을의 해석 단계에서는 Gadamer (1976)의 해석의 방법론과 함께 진행되었다. 말레이마을을 해석하는데 내부자적 관점에서 말레이 마을 (kampung)의 공간적 배치와 주택을 비롯한 다양한 장소에 부여된 의미를 이해해야 한다. 따라서 말레이마을의 요소를 다음과 같이 분류하였다. 1. 배경 (거시적 분석) 2. 마을주민 (거시적 분석) 3. 전반적 말레이 마을 (kampung) 배치 (거시적 분석) 4. 말레이 주택 (미시적 분석) 5. 말레이 주택정원 (미시적 분석) 6. 과수원과 숲 (미시적 분석) 7. 구불구불한 길 (미시적 분석) 8. 강 (미시적 분석) 9. 역사적, 문화적 요소 (미시적 분석) 이와 같은 파당창카드 마을의 구성요소를 통해 말레이마을 문화경관을 다양한 각도로 분석했다. 특히 이 요소들은 전반적으로 대부분의 말레이마을을 구성하며 외부인들은 각 개별 요소로 분류하기에는 난해하다. 각 요소를 통해 말레이 마을을 해석한 결과는 다음과 같다. 첫째, 페락 주(Perak State)에 위치한 파당창카드의 배경은 이 지역의 페락 왕(Perak Sultanate)으로 인하여 마을이 번창하고, 전통 수작업이 활성화되는 등 왕가와 밀접하게 관련된다. 둘째, 파당창카드의 주민들은 100년 이상을 거주하면서 자연과 사회를 이해하고, 적응하여 그들만의 환경과 문화가 공존한다. 셋째, 파당창카드의 배치는 기본적으로 원형의 보안림, 거대한 과수원이 강으로 에워싸는 형태이다. 그 중심에는 말레이 주택들이 위치하여 구불구불한 길이 배치되고, 식용식물과 유실수가 동시에 심겨져 형성된 마을이다. 넷째, 말레이의 삶과 전통을 가장 자연스럽게 보여주는 말레이 주택에서 주민들의 일상뿐만 아니라 그들의 정신과 삶의 모습을 볼 수 있다. 다섯째, 말레이 주택정원과 과수원, 그리고 숲은 말레이 주거환경의 경관을 반영한 곳으로 지역 주민들의 삶을 충족시킬 뿐만 아니라 말레이만의 전통 식재경관을 나타낸다. 여섯째, 구불구불한 길은 많은 사람들이 통과하는 길이자, 사교 모임의 장소로서의 역할도 수행한다. 일곱째, 파당창카드의 가장자리에 위치한 페락 강(Perak River)은 과거에 정치적, 경제적 수단이었으며, 현재에는 역사적, 사회문화적 요소로서 다양한 가치를 지닌 자연요소라 할 수 있다. 마지막으로 파당창카드의 역사적, 문화적 요소는 이 마을의 문화경관을 보강하면서 마을의 정체성을 보여주기도 하는 등 다양한 기능을 가지고 있다. 본 연구는 파당창카드의 각종 물리적 요소, 장소, 사람들, 그리고 이들의 활동을 토대로 말레이 마을의 문화경관의 의미를 해석하고자 했다. 파당창카드를 해석하면서 요소와 장소, 사람들 그리고 활동이 상호적으로 어떻게 엮이고, 경관을 형성하는지 파악하고자 한다. 결과적으로 본 연구는 파당창카드의 해석을 통해 도출한 결론을 통해 발전 방안을 제안하고자 한다. 파당창카드 마을의 발전 방안은 전통 문화를 유지하면서 전통 민속 마을로 선정하여 현 시대의 기술력을 동원하여 이 마을만의 문화경관을 유지하는 것이다. 이는 문화를 보존하는데 있어 시대의 변화에 맞춰 현대의 기술이 함께 적용되어야 한다는 팸턴(Fampton, 1983)의 맥락과 일치한다. 끝으로 한 특정마을이 성장할 수 있도록 설계하고 계획하기 위해서는 파당창카드를 통해 말레이 마을을 이해하고 문화경관을 해석하는 연구의 의미가 크다. 다시 말해, 본 연구는 앞으로 말레이시아에서 전통마을의 문화 경관이 보존되고 발전하는데 첫 걸음이 될 것이다.

The role of the glass transition in extruded cereal snack character

Teoh, Heidi Marie University of Illinois at Urbana-Champaign 2000 해외박사(DDOD)

This research was an investigation of the role the glass transition of cereal melts play in the determination of direct expanded extruded cornmeal snack character. The investigation began with the successful development of a thermal analysis method to identify glass transitions in extruded cornmeal puffs at cereal melt moistures (approximately 19%db). Isotherm and DSC data resulting from a study of the plasticization of water on Tg and texture indicated that extruded cornmeal puffs undergo glass transition at 24°C at a water activity of 0.87, corresponding to a moisture content of 20.8%db. Experimental texture results of the samples supported the isotherm and DSC data used to determine the water activity and moisture content sufficient to depress Tg to room temperature. Investigation of cornmeal components utilized dynamic vapor sorption (DVS), a new technology, to generate sorption isotherms and rapidly condition samples. Component isotherm data allowed for the accurate prediction of cornmeal sorption behavior. As expected, cornmeal isotherm and Tg behavior was similar to the behavior of native dent starch. Analysis of data indicated that the components contributed to cornmeal Tg values on a weighted average basis. Cornmeal was extruded in a 4 x 2 factorial design to produce samples covering a range of product character. The extrudates were characterized by extruder response, physical properties, texture properties, and polymeric properties. As the severity of the extrusion scenario increased, expansion properties increased and the texture properties were more acceptable for this type of product. As SME values increased, both molecular weight and Tg values decreased in agreement with glass transition theory. It was proposed that the glass transition plays a role in the development of product structure as a result of its relationship to melt viscosity. It was proposed here that the location of Tg relative to the cereal melt temperatures (ΔT) determines the viscosity of the melts (High ΔT equates to low melt viscosity). The viscosity of the cereal melt controls the rate and extent of expansion that occurs upon expulsion from the die. Experimental results supported the proposed mechanisms of effect of the glass transition on cornmeal snack character.

Design of Hybrid Passive and Active Mechanisms for Control of Insect-Scale Flapping-Wing Robots

Teoh, Zhi Ern Harvard University ProQuest Dissertations & Theses 2015 해외박사(DDOD)

Flying insects exhibit a remarkable ability to fly in environments that are small, cluttered and highly dynamic. Inspired by these animals, scientist have made great strides in understanding the aerodynamic mechanisms behind insect-scale flapping-wing flight. By applying these mechanisms together with recent advances in meso-scale fabrication techniques, engineers built an insect-scale flapping-wing robot and demonstrated hover by actively controlling the robot about its roll and pitch axes. The robot, however, lacked control over its yaw axis preventing control over its heading angle. In this thesis, we show that the roll and pitch axes of a single actuator insect-scale flapping-wing robot can also be passively stabilized by the addition of a pair of aerodynamic dampers. We develop design guidelines for these dampers, showing that the previously unstable robot with the addition of the dampers is able to perform stable vertical flights and altitude control. To address the lack of yaw control, we develop a yaw torque generating mechanism inspired by the fruit fly wing hinge. We present the development of this mechanism in three stages: from the conceptual stage, to the torque measurement stage and finally to a hover capable stage. We show that the robot is able to generate sufficient yaw torque enabling the robot to transition from hover to heading control maneuvers.

A girl without talent is therefore virtuous: Educating Chinese women in British Malaya and Singapore, 1850s--1960s

Teoh, Karen May-Shen Harvard University 2008 해외박사(DDOD)

The history of English- and Chinese-medium girls' schools that overseas Chinese founded and attended from the late nineteenth to the mid-twentieth centuries in colonial Malaya and Singapore reveals the conflicts between constructions of modernization and ethnic Chinese identity, and how female education, as an integral component of both processes, created situations in which society and individuals have had to grapple with the challenge of resolving these contradictions within themselves. Using a combination of archival materials and oral histories, this study takes a cross-cultural and comparative perspective, looking into the motivations and strategies of various groups (European missionaries, colonial authorities and Chinese nationalists) for educating girls, and the influence that this education had on the formation of ethnic identities among overseas Chinese women. To the various authorities involved, managing the female population through formal training fit into the vital agenda of propelling society towards modernity, even as the definition of "modern" varied across time and community. At the same time, girls' schools continued to stress the importance of what they perceived as significant feminine virtues. Chinese girls were being sent out of the home to be equipped with new knowledge, work skills, cultural orientations and political awareness, only to be brought back into the home to fulfill their roles as household managers, teachers of their children, and, at least in theory, guardians of cultural traditions. As key venues for the emergence of tension between modernization and cultural conservatism within the Chinese diasporic community, girls' schools are a lens through which we can view the manifestation of a common anxiety: the potential loss of traditional values and ethnic authenticity in a rapidly globalizing environment. Up to and beyond the period of decolonization in the late 1950s, transnational forces such as British imperialism and Chinese nationalism pulled Chinese migrants in disparate directions. For overseas Chinese women---a population thrice marginalized by its gender, ethnicity and migrant status in a European colony---the competing demands of these diverse forces created a range of ways in which they formulated their identities as women, as Chinese, and as part of a transnational diaspora.

Error Detection in Bosonic Circuit Quantum Electrodynamics

Teoh, James David Yale University ProQuest Dissertations & Theses 2023 해외박사(DDOD)

Any useful quantum computer must be robust to errors on the individual qubits. This means a quantum algorithm can still reliably give the correct answer despite the qubits suffering from decoherence. Quantum error correction provides a means for achieving this, by encoding logical qubits redundantly in many additional physical qubits and/or additional energy levels. In the presence of errors the logical information is still preserved, but this remains a monumentally difficult task to achieve in practice. But what can we do with quantum error detection? Rather than trying to recover from a quantum error we instead reset qubits that are known to have suffered errors. If we know exactly when and which qubit suffered an error, this becomes an erasure, which is significantly easier to correct than Pauli errors. For erasure qubits to be viable, we must be able to detect errors that happen not just while idling as for a quantum memory, but during every operation needed to run a quantum computer. In this thesis we describe a way to realize the necessary error-detected operations in the qubit platform of bosonic circuit quantum electrodynamics. We choose to encode our qubits in 3D superconducting microwave cavity modes, and introduce the dual-rail cavity qubit, in which the dominant hardware errors can all be detected as erasure errors. With this new approach, practical quantum error correction is expected to be within reach for currently available experimental hardware and coherence times. Finally, we also investigate how to use error detection to circumvent loss errors in links that connect qubit modules together, experimentally realizing high-fidelity state transfer between modules.

Transforming growth factor beta-induced tyrosine phosphorylation of signal transducer and activator of transcription 3 and cellular invasion is mediated by interleukin-6 secretion in breast cancer cell lines

Parker, Nicole N. Teoh University of Florida 2009 해외박사(DDOD)

A hallmark of breast cancers is the overexpression or constitutive activation of various oncoproteins. Two examples are the proteins Signal Transducer and Activator of Transcription 3 (STAT3) and Transforming Growth Factor beta (TGFbeta). Although each of these signaling molecules and their respective pathways have been implicated in cancer development, the mechanism of cross-signaling between these pathways has not been established in the cancer setting. We hypothesize that cross-talk between the TGFbeta and STAT3 pathways contributes to cancer invasiveness through an autocrine signaling loop. Preliminary data in nontransformed mouse mammary NMuMG cells indicate that exogenous TGFbeta treatment results in phosphorylation of STAT3 on its activating Tyrosine (705) site. Strikingly, this effect is observed only after several hours of TGFbeta treatment and appears to be mediated by the cytokine Interleukin-6 (IL-6). TGFbeta induced IL-6 mRNA upregulation occurs concomitantly with TGFbeta-stimulated STAT3 tyrosine phosphorylation. Blockade of IL-6 function with a mouse IL-6 receptor fusion protein abrogates this effect, thereby implicating IL-6 as the factor responsible for TGFbeta-stimulated STAT3 tyrosine phosphorylation in NMuMG cells. TGFbeta confers an invasive phenotype on NMuMG and Mv1Lu cells that is IL-6 dependent. Examination of this mechanism in human breast cancer cell lines suggests that TGFbeta-mediated secretion of IL-6 is responsible for the constitutive STAT3 tyrosine phosphorylation observed in many breast cancer cell lines. The invasive human MDA-MB-231 basal-like breast cancer cell line exhibits TGFbeta and IL-6 dependent invasion, while the noninvasive human MDA-MB-361 luminal breast cancer cell line is capable of attaining IL-6-dependent cellular invasiveness in response to TGFbeta treatment. (Full text of this dissertation may be available via the University of Florida Libraries web site. Please check http://www.uflib.ufl.edu/etd.html).