Search for: All records

Models recently used in the literature proving residual networks (ResNets) are better than linear predictors are actually different from standard ResNets that have been widely used in computer vision. In addition to the assumptions such as scalar-valued output or single residual block, the models fundamentally considered in the literature have no nonlinearities at the final residual representation that feeds into the final affine layer. To codify such a difference in nonlinearities and reveal a linear estimation property, we define ResNEsts, i.e., Residual Nonlinear Estimators, by simply dropping nonlinearities at the last residual representation from standard ResNets. We show that wide ResNEsts with bottleneck blocks can always guarantee a very desirable training property that standard ResNets aim to achieve, i.e., adding more blocks does not decrease performance given the same set of basis elements. To prove that, we first recognize ResNEsts are basis function models that are limited by a coupling problem in basis learning and linear prediction. Then, to decouple prediction weights from basis learning, we construct a special architecture termed augmented ResNEst (A-ResNEst) that always guarantees no worse performance with the addition of a block. As a result, such an A-ResNEst establishes empirical risk lower bounds for a ResNEst using corresponding bases. Our results demonstrate ResNEsts indeed have a problem of diminishing feature reuse; however, it can be avoided by sufficiently expanding or widening the input space, leading to the above-mentioned desirable property. Inspired by the densely connected networks (DenseNets) that have been shown to outperform ResNets, we also propose a corresponding new model called Densely connected Nonlinear Estimator (DenseNEst). We show that any DenseNEst can be represented as a wide ResNEst with bottleneck blocks. Unlike ResNEsts, DenseNEsts exhibit the desirable property without any special = architectural re-design. 

Over the past few decades, rapid development of laser cooling techniques and narrow-linewidth lasers have allowed atom-based quantum clocks to achieve unprecedented precision. Techniques originally developed for atomic clocks can be extended to ultracold molecules, with applications ranging from quantum-state-controlled ultracold chemistry to searches for new physics. Because of the richness of molecular structure, quantum metrology based on molecules provides possibilities for testing physics that is beyond the scope of traditional atomic clocks. This thesis presents the work performed to establish a state-of-the-art quantum clock based on ultracold molecules. The molecular clock is based on a frequency difference between two vibrational levels in the electronic ground state of 88Sr2 diatomic molecules. Such a clock allows us test molecular QED, improve constraints on nanometer-scale gravity, and potentially provide a model-independent test of temporal variations of the proton-electron mass ratio. Trap-insensitive spectroscopy is crucial for extending coherent molecule-light interactions and achieving a high quality factor Q. We have demonstrated a magic wavelength technique for molecules by manipulating the optical lattice frequency near narrow polarizability resonances. This general technique allows us to increase the coherence time to tens of ms, an improvement of a factor of several thousand, and to narrow the linewidth of a 25 THz vibrational transition initially to 30 Hz. This width corresponds to the quality factor Q = 8 × 10^11. Besides the molecular quantum metrology, investigations of novel phenomena in state-selected photodissociation are also described in this thesis, including magnetic-field control of photodissociation and observation of the crossover from ultracold to quasiclassical chemistry. 

Gaia16aye was a binary microlensing event discovered in the direction towards the northern Galactic disc and was one of the first microlensing events detected and alerted to by the Gaia space mission. Its light curve exhibited five distinct brightening episodes, reaching up to I  = 12 mag, and it was covered in great detail with almost 25 000 data points gathered by a network of telescopes. We present the photometric and spectroscopic follow-up covering 500 days of the event evolution. We employed a full Keplerian binary orbit microlensing model combined with the motion of Earth and Gaia around the Sun to reproduce the complex light curve. The photometric data allowed us to solve the microlensing event entirely and to derive the complete and unique set of orbital parameters of the binary lensing system. We also report on the detection of the first-ever microlensing space-parallax between the Earth and Gaia located at L2. The properties of the binary system were derived from microlensing parameters, and we found that the system is composed of two main-sequence stars with masses 0.57 ± 0.05 M ⊙ and 0.36 ± 0.03 M ⊙ at 780 pc, with an orbital period of 2.88 years and an eccentricity of 0.30. We also predict the astrometric microlensing signal for this binary lens as it will be seen by Gaia as well as the radial velocity curve for the binary system. Events such as Gaia16aye indicate the potential for the microlensing method of probing the mass function of dark objects, including black holes, in directions other than that of the Galactic bulge. This case also emphasises the importance of long-term time-domain coordinated observations that can be made with a network of heterogeneous telescopes.