Graduate School in Computer Science at Tufts University, starting Fall 2016
Dear Student,
Greetings from Tufts University, in the greater Boston area!
I am writing to let you know that the Department of Computer Science is looking for exceptional students interested in pursuing PhDs or MS degrees.
The Department research is vibrant (see snippets of sample research works at the bottom of this email), and has expertise in several areas including:
· Computational & Systems Biology
· Cognitive Science
· Human-Computer Interaction
· Networking, Mobile, and Cloud Computing
· Machine Learning and Data Mining
· Programming Languages and Systems
· Robotics and Human-Robot Interaction
· Analytics & Visualization
The department has 20 faculty, and about 60 full-time graduate students. The department offers transformative student experiences, and fosters a collegial and nurturing environment. Students enjoy close interactions with the faculty. The campus is beautiful. Living in Boston is a great life experience. We have many international students. Students are successfully employed post-graduation in academic jobs (Wellesley, Northwestern, Baylor, Carleton, Drexel, and UT Austin, DePaul, and Smith), as post-docs (MIT, Harvard, UMass Medical School), and in high tech companies such as amazon and google and startups.
For 2015-2016, we will have several open teaching and research assistantship positions. These positions include tuition waivers, and a stipend that covers living expenses. These positions are typically made available for those pursuing a PhD degree.
Applicants should be among the top 10-15% within his or her graduating class. GRE and are required. TOEFL scores may be required. Applicants with a master's degree and/or industrial experience are strongly encouraged to apply.
I encourage you to apply! The deadline is Jan 15, 2016.
If you have any questions, please do not hesitate to email (subject heading: "Fall 2016: Potential Graduate Student: your name")
To apply, visit:
http://www.cs.tufts.edu/Academics/applying-to-the-phd-program.html
The online application is here:
http://gradstudy.tufts.edu/admissions/howtoapply.htm
Research Highlights
Slonim and her collaborators, including Dr. Diana Bianchi from Tufts Medical School, addressed the problem ofcharacterizing systematic anomalies in expression data. The challenge was identifying “abnormal” (diseased) samples given only “normal” (healthy) samples as training set. This problem is difficult because expression data is noisy and has high dimensionality. Slonim's method not only determines outlier samples, but also “explains” them, by identifying the gene sets that are most responsible for the anomalous behavior. Slonim applies this method to study the impact of maternal obesity on fetal development. J. of Computational Biology (2014)
Dogar aims to create seamless experiences when using interactive applications (e.g., voice, video conferencing, online gaming, etc). Dogar investigates how to efficiently combine cheap but unreliable paths in today's Internet with the highly reliable but expensive cloud paths (owned by companies like Microsoft and Google). The key idea is to send a small number of "special" packets on these cloud paths. In case of a packet loss on Internet paths, these special packets are used to "reconstruct" the lost packet. This work will be featured in the SIGCOMM HotNets workshop, the premier venue for transformative ideas for the networking community. SIGCOMM HotNets (2015)
Machine learning solves a variety of prediction problems, including classification (predicting if a credit card transaction fraudulent), regression (predicting the amount of revenue from a new customer), and count prediction (predicting the number of customers wishing to buy a new product). Bayesian non-parametric models offer a promising approach to such problems in that they provide accurate predictions but they are typically too slow for large scale datasets. To address this, Khardon and PhD student Sheth have developed a "generic sparse Gaussian Process approximate inference algorithm", a single efficient approximation algorithm that applies across all such prediction problems. The significance is in expanding the range of applicability of non-parametric methods, and in avoiding the need for ad hoc solutions for each type of prediction problem. International Conference on Machine Learning (2015)
Cowen and Hescott and their research students designed a novel method to incorporate knowledge of known biological pathways into metrics of protein relationships in the interactome. The network of protein-protein interactions, in effect the "Facebook friend" graph of which proteins hang out together in the cell, can be mined for information that allows the prediction of protein function based on "guilt by association" coupled with the function of proteins in a protein's local neighborhood. In the setting where locality is defined in these networks using a diffusion process, they showed that known biological pathways could be hardwired into the diffusion process. This allowed them to leverage both the pairwise interaction information that was available for most proteins, and the very detailed pathway knowledge that is available for some medically important proteins at the same time, improving the state of the art in protein function prediction based on biological network data. Bioinformatics (2014)
Best,
soha
Soha Hassoun
Professor and Chair
Department of Computer Science
Tufts University
http://www.cs.tufts.edu/~soha/
Greetings from Tufts University, in the greater Boston area!
I am writing to let you know that the Department of Computer Science is looking for exceptional students interested in pursuing PhDs or MS degrees.
The Department research is vibrant (see snippets of sample research works at the bottom of this email), and has expertise in several areas including:
· Computational & Systems Biology
· Cognitive Science
· Human-Computer Interaction
· Networking, Mobile, and Cloud Computing
· Machine Learning and Data Mining
· Programming Languages and Systems
· Robotics and Human-Robot Interaction
· Analytics & Visualization
The department has 20 faculty, and about 60 full-time graduate students. The department offers transformative student experiences, and fosters a collegial and nurturing environment. Students enjoy close interactions with the faculty. The campus is beautiful. Living in Boston is a great life experience. We have many international students. Students are successfully employed post-graduation in academic jobs (Wellesley, Northwestern, Baylor, Carleton, Drexel, and UT Austin, DePaul, and Smith), as post-docs (MIT, Harvard, UMass Medical School), and in high tech companies such as amazon and google and startups.
For 2015-2016, we will have several open teaching and research assistantship positions. These positions include tuition waivers, and a stipend that covers living expenses. These positions are typically made available for those pursuing a PhD degree.
Applicants should be among the top 10-15% within his or her graduating class. GRE and are required. TOEFL scores may be required. Applicants with a master's degree and/or industrial experience are strongly encouraged to apply.
I encourage you to apply! The deadline is Jan 15, 2016.
If you have any questions, please do not hesitate to email (subject heading: "Fall 2016: Potential Graduate Student: your name")
To apply, visit:
http://www.cs.tufts.edu/Academics/applying-to-the-phd-program.html
The online application is here:
http://gradstudy.tufts.edu/admissions/howtoapply.htm
Research Highlights
Slonim and her collaborators, including Dr. Diana Bianchi from Tufts Medical School, addressed the problem ofcharacterizing systematic anomalies in expression data. The challenge was identifying “abnormal” (diseased) samples given only “normal” (healthy) samples as training set. This problem is difficult because expression data is noisy and has high dimensionality. Slonim's method not only determines outlier samples, but also “explains” them, by identifying the gene sets that are most responsible for the anomalous behavior. Slonim applies this method to study the impact of maternal obesity on fetal development. J. of Computational Biology (2014)
Dogar aims to create seamless experiences when using interactive applications (e.g., voice, video conferencing, online gaming, etc). Dogar investigates how to efficiently combine cheap but unreliable paths in today's Internet with the highly reliable but expensive cloud paths (owned by companies like Microsoft and Google). The key idea is to send a small number of "special" packets on these cloud paths. In case of a packet loss on Internet paths, these special packets are used to "reconstruct" the lost packet. This work will be featured in the SIGCOMM HotNets workshop, the premier venue for transformative ideas for the networking community. SIGCOMM HotNets (2015)
Machine learning solves a variety of prediction problems, including classification (predicting if a credit card transaction fraudulent), regression (predicting the amount of revenue from a new customer), and count prediction (predicting the number of customers wishing to buy a new product). Bayesian non-parametric models offer a promising approach to such problems in that they provide accurate predictions but they are typically too slow for large scale datasets. To address this, Khardon and PhD student Sheth have developed a "generic sparse Gaussian Process approximate inference algorithm", a single efficient approximation algorithm that applies across all such prediction problems. The significance is in expanding the range of applicability of non-parametric methods, and in avoiding the need for ad hoc solutions for each type of prediction problem. International Conference on Machine Learning (2015)
Cowen and Hescott and their research students designed a novel method to incorporate knowledge of known biological pathways into metrics of protein relationships in the interactome. The network of protein-protein interactions, in effect the "Facebook friend" graph of which proteins hang out together in the cell, can be mined for information that allows the prediction of protein function based on "guilt by association" coupled with the function of proteins in a protein's local neighborhood. In the setting where locality is defined in these networks using a diffusion process, they showed that known biological pathways could be hardwired into the diffusion process. This allowed them to leverage both the pairwise interaction information that was available for most proteins, and the very detailed pathway knowledge that is available for some medically important proteins at the same time, improving the state of the art in protein function prediction based on biological network data. Bioinformatics (2014)
Best,
soha
Soha Hassoun
Professor and Chair
Department of Computer Science
Tufts University
http://www.cs.tufts.edu/~soha/
瀏覽數: