136879 research outputs found
Sort by
Machine Learning for CMS Online Muon Segment Reconstruction
No full textThe Large Hadron Collider (LHC) collides protons at near the speed of light to shed light on new physics beyond the Standard Model. As the LHC undergoes increases in luminosity, the amount of data that has to be processed will increase by up to a factor of ten. As the latency requirements and sheer size of data generated dictate that not all data can be processed or kept, multi-stage trigger algorithms decide what data is signs of interesting physics. New detectors and higher data throughput necessitate upgrades in these trigger algorithms. Machine learning is a promising technique that has begun to be used especially in the higher levels of triggering and analysis, but is not currently implemented in some of the lowest levels of triggering. The Compact Muon Solenoid (CMS) is one of the large particle physics experiments on the LHC ring, and it is composed of multiple detector systems that serve different roles. ME0 is the innermost muon detector system, composed of six interlinked detector layers. We investigate machine learning for muon segment and transverse momentum assignment in ME0. A small, fully connected network discriminates between background noise and muon signal with high accuracy. This model was pared down from larger sizes, and accuracy was maintained while reducing over-fitting and having closer connection between accuracy and validation accuracy. The model was compared with the current algorithm using an emulator. The novel implementation had 98.3% efficiency and 11.4% purity, whereas the current implementation has 99.0% efficiency and 18.9% purity. CMS has a strong axial magnetic field, which makes charged particles such as muons bend proportionally to their transverse momentum. Therefore, determining the bend angle of a particle track provides a good measurement of its momentum. We also investigate assignment of transverse momentum to muons using machine learning. Preliminary models failed to train well, so more analysis of data was conducted. A two mode distribution of transverse momentum was found with the cutoff being around 0.3 MeV. With this new knowledge, experiments still largely failed to improve. Future work will be necessary to better understand both segment identification and transverse momentum assignment and classification. However, this is a start for potential benefits of applying machine learning earlier in the trigger path and presents a foundation for future research
Encapsulation of Water-Sensitive Compositions Using Non-Aqueous Pickering Emulsions
No full textEncapsulation of water-sensitive compounds like ionic liquids (ILs) or salt hydrate phase change materials (PCMs) can address limitations associated with the bulk, while facilitating their handling. For example, encapsulation can increase the active surface area of the liquid and mitigate slow mass transfer rates caused by high viscosities, or the shell can support surface-initiated crystallization. Further, a protective shell around the core material can overcome its susceptibility to the external environment. This dissertation reports soft-templated encapsulation strategies for water-sensitive or water-miscible compositions, such as ILs and salt hydrate PCMs by leveraging non-aqueous Pickering emulsions. This work establishes that control of the composition of emulsions is critical for encapsulating hydrophilic or hygroscopic compounds.
Capsules of novel fluorinated ionic liquids (FILs) are architected with composite graphene oxide (GO)/polymer shells by employing interfacial polymerization in non-aqueous Pickering emulsions, stabilized by alkylated GO nanosheets. Here, interfacial polymerization occurs between a multifunctional monomer present in the dispersed phase of the emulsion and a complementary monomer introduced to the continuous phase. While capsules can be readily formed, this approach does not yield a pristine core, as the core liquid contains residual unreacted monomers and other shell-forming precursors. Contamination of the capsule cores can have a significant impact on salt hydrate PCMs as their thermal energy storage characteristics are highly dependent on their purity and retention of stoichiometry. To maintain core integrity, alternative soft-templated encapsulation techniques were examined, specifically those that do not rely on interfacial polymerization for shell formation. This entailed in-situ polymerization of monomers within the continuous phase of non-aqueous Pickering emulsions, external crosslinking of reactive particle surfactants at the fluid-fluid interface, and precipitation of commodity polymers onto emulsion droplets. The latter single-step methodology led to the formation of microcapsules loaded with pristine salt hydrate or IL cores that exhibit structural integrity and stability to solid-liquid phase change cycles. Additionally, printing-induced encapsulation is demonstrated by direct ink writing of shear-thinning and thixotropic inks. Salt hydrate particles, derived from non-aqueous Pickering emulsions, are utilized as rheological modifiers to formulate these inks, enabling the printing of non-corrosive salt hydrate PCM composites.
Collectively, the work reported in this thesis represents notable advancements in the encapsulation of compositionally sensitive materials, tailoring hybrid structures with promising applications in areas such as gas uptake and thermal energy storage. Further, the tunability of the properties in the resulting composites allows for a better delineation of structure-processing-property-application relationships, as relevant to various technologies
Causal Inference with Differential Privacy
No full textIn the social and health sciences, researchers often make causal inferences using sensitive variables.
These researchers, as well as the data holders themselves, may be ethically and perhaps legally obligated to protect the confidentiality of study participants' data. It is now known that releasing any statistics, including estimates of causal effects, computed with confidential data leaks information about the underlying data values. Thus, analysts may desire to use causal estimators that can provably bound this information leakage.
Motivated by this goal, we develop algorithms for estimating weighted average treatment effects with binary outcomes that satisfy the criterion of differential privacy.
We present theoretical results on the accuracy of several differentially private estimators of weighted average treatment effects. We illustrate the empirical performance of these estimators using simulated data and a causal analysis using data on education and income