|
|
| Name : Haijun Yang
Specialty : Experimental Particle Physics Phone : 1-734-763-2329 (Office), 1-734-763-7208 (Lab) E-mail : yhj@umich.edu Website : at UMICH Publications: Selected Publications Talks : Recent Presentations |
|
|
|
Aug.2005 -- Present
Oct.2001 -- Jul.2005
Aug. 2000 -- Oct. 2001
|
|
|
|
July.1999 -- July.2000
July.1998 -- July.1999
Mar. 1997 -- July.1998
Sept.1995 -- Mar. 1997
Sept.1991 -- July. 1995 Dept. of Physics, Zhejiang University.(BS) Sept.1988 -- July. 1991 Wenling High School. Sept.1985 -- July. 1988 Chuanbei Junior High School. Sept.1980 -- July. 1985 Shengbei Elementary School. |
|
|
|
|
|
|
|
I have been involved in three projects, L3 and CERN, R\&D of International Linear Collider (ILC)
and Booster Neutrino Experiment (MiniBooNE) at Fermilab over the past five years. I'm a Principal
Investigator (PI) of DOE/LANL grant for MiniBooNE event reconstruction, particle identification
and neutrino oscillation analysis at University of Michigan since October, 2005.
==> L3 at CERN <== I have been working with Prof. Keith Riles (together with other L3 collaborators) on a search for Minimal Supersymmetric Standard Model (MSSM) neutral Higgs bosons and Gauge Mediated Supersymmetric Breaking (GMSB) particles using data collected by L3 detector at LEPII from August, 2000 through the end of 2003. No evidence for MSSM neutral Higgs bosons is found and large regions of parameter spaces are excluded; the lower limits on Higgs boson massess are derived at the 95\% confidence level to be $m_h > 84.5$ GeV and $m_A > 86.3$ GeV. These results are published in Physics Letters B 545(2002). The results of the search for GMSB particles are described in L3 note 2777. ==> R&D of International Linear Collider <== At the same period of time, I spent part time on the next generation $e^+e^-$ linear collider (ILC) detector project with Prof. Riles. Our specific interests include physics and detector simulations for measurement of Higgs and SUSY properties, for optimization of the tracking system, and for the design of a high-precision optical alignment system using Frequency Scanned Interferometry( FSI) for the international linear collider silicon tracker detector. The center-of-mass energies of the linear collider will range from 350 GeV to 1000 GeV, enabling precise standard model measurements and study of new particles which may include the Higgs bosons and supersymmetric particles. Our group has constructed several demonstration Frequency Scanned Interferometer systems with the laser light transported by air or single-mode optical fiber, using single-laser and dual-laser scanning techniques for initial feasibility studies. Absolute distance was determined by counting the interference fringes produced while scanning the laser frequency. The main goal of the demonstration systems was to determine the potential accuracy of absolute distance measurements that could be achieved under both controlled and realistic conditions. Secondary goals included estimating the effects of vibrations and studying error sources crucial to the absolute distance accuracy. Two multiple-distance-measurement analysis techniques were developed to improve distance precision and to extract the amplitude and frequency of vibrations. Under well controlled laboratory conditions, a measurement precision of $\sim$ 50 nm was achieved for absolute distances ranging from 0.1 meters to 0.7 meters by using the first multiple-distance-measurement technique (slip measurement window with fixed size). The second analysis technique (slip measurement window with fixed start point) has the capability to measure vibration frequencies ranging from 0.1 Hz to 100 Hz with amplitude as small as a few nanometers, without a {\em priori} knowledge. The multiple-distance-measurement analysis techniques are well suited for reducing vibration effects and uncertainties from fringe \& frequency determination, but do not handle well the drift errors such as from thermal effects. This work, ``High-precision Absolute Distance and Vibration Measurement by Using Frequency Scanned Interferometry'', has been published in Applied Optics, Vol.44, No.19:3937-3944, 2005. The dual-laser scanning technique was pioneered by the Oxford group for alignment of the ATLAS Semi-conductor tracker; it was demonstrated that precisions of better than 0.4 $\mu$m and 0.25 $\mu$m for distances of 0.4 m and 1.195 m, respectively. In our recent studies, we combine our multi-distance-measurement analysis technique (slip measurement window with fixed size) with the dual-laser scanning technique improve the absolute distance measurement precision. The multi-distance-measurement technique is effective in reducing uncertainties from vibration effects and interference fringe determination, while the dual-laser scanning allows for cancellation of drift errors. A precision better than 0.2 microns was achieved for a distance of 0.41 meters under realistic conditions, even with fan and vibration source. The 2nd paper, ``High-precision Absolute Distance Measurment using Dual-Laser Frequency Scanned Interferometry Under Realistic Conditions'', is under preparation and will be submitted to Applied Optics shortly. This work has been supported by NSF and DOE in recent years. I would like to continue efforts with Prof. Riles and student(s) on the international linear collider detector R\&D in the coming years towards the construction and operation of the linear collider. ==> MiniBooNE Experiment at Fermilab <== Since November, 2003, I have been involved (part time) in particle identification (PID) for the MiniBooNE experiment at the Fermi National Accelerator Laboratory which is designed to confirm or refute the evidence for $\nu_\mu \rightarrow \nu_e$ oscillations at $\Delta m^2 \sim 1~eV^2/c^4$ seen by the LSND experiment. It is a crucial experiment which will imply new physics beyond the standard model if the LSND signal is confirmed. Together with Prof. Ji Zhu (Department of Statistics) and Prof. Byron P. Roe, I have applied the boosting algorithm to the MiniBooNE PID to classify the events. It turns out that PID with the boosting algorithm is 20\% to 80\% better than that with our standard artificial neural network(ANN) PID technique. Boosting performance relative to that of an ANN depends on the Monte Carlo samples and PID variables. As far as we know, it is the first time the boosting algorithm has been applied to a particle physics experiment. One paper (physics/0408124), "Boosted Decision Trees as an Alternative to Artificial Neural Networks for Particle Identification", was published by Nuclear Instruments \& Methods in Physics Research Section A in May, 2005. The Michigan group had the leading role on the paper and I'm the corresponding author. Although the boosting algorithm was tested in only one experiment, it is anticipated to have wide application in physics, especially in data analysis of high energy physics experiments for signal and background events separation. About a dozen people from several major HEP experiments (MiniBooNE, CDF, D0, BaBar, ATLAS, CMS etc.) have asked us for further information and source codes for the boosting algorithms since our first report on the boosting PID performance. More comprehensive studies of boosted decision trees with various boosting algorithms for MiniBooNE particle identification are also published in Nucl. Instrum. \& Meth. A555:370-385, 2005, I'm the corresponding author of the paper. The motivation for the boosting algorithm is to design a procedure that combines many ``weak'' classifiers (such as decision trees, random forests, ANNs etc.) to achieve a powerful classifier. One starts with unweighted training events and builds a decision tree. If a training event is misclassified, then the weight of that event is increased (boosted). A second tree is built using exactly the same set of training events but with new weights. Again misclassified events have their weights boosted and the procedure is repeated several hundred to thousand times until the performance reaches optimal. For a given test event, it is followed through each tree in turn, if it lands on the signal leaf it is given a score of 1, otherwise -1. The sum of scores from all trees (possibly weighted), is the final score of the event. High score means the event is most likely signal and low score, background. The major advantages of boosted decision trees are their stability, their ability to handle large number of input variables, and their use of boosted weights for misclassified events to give these events a better chance to be correctly classified in succeeding trees. Besides the significant improvement of the MiniBooNE PID performance by using boosted decision trees, I also played key role in exploring over 1000 new possible variables and selecting dozens of the most powerful variables to make further PID improvement. We obtained about 50\% additional reduction of background events while keeping the same signal efficiency. I presented and tried to suppress one of the major backgroud sources, neutral current radiative $\Delta \rightarrow N + \gamma$, by boosting training. Many colleagues thought it would be very difficult to suppress this background since the photon of the background will produce a single electron-like ring, an irreducible background to the oscillation $\nu_e$ signal. The boosted decision trees can make full use of many small differences between the signal and background rings to produce powerful separation; the radiative $\Delta$ background was reduced about 40\% by using boosting training. Furthermore, I combined reconstructed variables from two parallel reconstruction packages for the boosting training (named YBoosts). The signal efficiency improved an additional 20\% for the YBoosts while keeping the same background contamination ratio. The overall performance of the YBoosts (included the stability, agreement between data and MC events and PID efficiency) are the best obtained for the MiniBooNE experiment so far. All of the above improvements were cross checked and verified by other groups within the MiniBooNE Collaboration. For the March 2005 baseline, only one and half years after I became involved in the MiniBooNE experiment, the $\nu_e$ signal efficiency has increased by a factor of at least 2 compared with Runplan 2003 results for the same background contamination ratio. The MiniBooNE experiment hopes, but has no guarantee, to obtain sufficient data ($10^{21}$ POT) to reach a definitive conclusion concerning the LSND result. The significant improvements obtained by all these efforts and efforts from all other collaborators enable us to reach a given similar neutrino oscillation sensitivity with fewer POT. It has a major impact for the MiniBooNE experiment. For the latest November 2005 baseline, I built six sets of YBoost PIDs using different sets of input variables selected by the Algorithm Group for training. Each set of YBoosts have 11 boosting outputs trained with various signal and background samples which are motivated to optimize physics results of wide range data analysis in the MiniBooNE experiment. Some examples are $\nu_\mu \rightarrow \nu_e$ oscillation analysis, $\nu_\mu$ disapperance analysis, the cross section or branching ratio measurement of $\nu_\mu$ charged current quasi elastic events (CCQE), neutral current $\pi^0$ events, charged current $\pi^0$ events, charged current $\pi^+$ events, neutral current radiative $\Delta$ events and charged current $\pi^+$ from $\nu_e$ etc. I have been supported by the Los Alamos National Laboratory for the work on the MiniBooNE experiment since October, 2004. I would like to continue effort on the MiniBooNE PID and data analysis in the coming years. What's more, I would like to spend part of my time on an ongoing experiment such as ATLAS at CERN with Prof. Bing Zhou, in order to apply some advanced data mining techniques, such as boosted decision trees and boosted random forests etc. which are tested using the MiniBooNE MC samples, to improve the PID performance and to prepare data analysis software for searches for the Higgs/SUSY particles in the coming LHC era. |
|
|
| # L3 international collaboration
at European Organization for Nuclear Research(CERN),
Geneva, Switzerland
# MiniBooNE collaboration at Fermi National Accelerator Laboratory(FNAL), Batavia, IL, USA # American Linear Collider Working Group(NLC) at Stanford Linear Accelerator Center(SLAC), Stanford University. # Tracking Group of the University Consortium for the Linear Collider(UCLC) # American Physical Society (APS), USA. |
|
|
| 1) Privilege to enter Ph.D Program at Institute of High Energy
Physics. Exempted from admission exam, 1997
2) Privilege to enter Graduate Program at Chinese Academy of Sciences, Exempted from admission exam, 1995 3) "Outstanding Graduate Student" award received from government of Zhejiang Province, 1995 4) "Guang Hua" Scholarship, 1994 (one of the highest honor in ZheJiang Univ.) 5) Excellent Student Scholarship, 1992, 1993, 1994 |
-- Revised by Haijun Yang at University of Michigan on February 3, 2006