arXiv subjects

Federico Berto

Publications and source records attributed to Federico Berto.

25 records · Page 2Linked to original sources

arXiv2023

DevFormer: A Symmetric Transformer for Context-Aware Device Placement

In this paper, we present DevFormer, a novel transformer-based architecture for addressing the complex and computationally demanding problem of hardware design optimization. Despite the demonstrated efficacy of transformers in domains including natural language processing and computer vision, their use in hardware design has been limited by the scarcity of offline data. Our approach addresses this limitation by introducing strong inductive biases such as relative positional embeddings and action-permutation symmetricity that effectively capture the hardware context and enable efficient design optimization with limited offline data. We apply DevFoemer to the problem of decoupling capacitor placement and show that it outperforms state-of-the-art methods in both simulated and real hardware, leading to improved performances while reducing the number of components by more than $30\%$. Finally, we show that our approach achieves promising results in other offline contextual learning-based combinatorial optimization tasks.

cs.LG↗

arXiv2022

Transform Once: Efficient Operator Learning in Frequency Domain

Spectral analysis provides one of the most effective paradigms for information-preserving dimensionality reduction, as simple descriptions of naturally occurring signals are often obtained via few terms of periodic basis functions. In this work, we study deep neural networks designed to harness the structure in frequency domain for efficient learning of long-range correlations in space or time: frequency-domain models (FDMs). Existing FDMs are based on complex-valued transforms i.e. Fourier Transforms (FT), and layers that perform computation on the spectrum and input data separately. This design introduces considerable computational overhead: for each layer, a forward and inverse FT. Instead, this work introduces a blueprint for frequency domain learning through a single transform: transform once (T1). To enable efficient, direct learning in the frequency domain we derive a variance-preserving weight initialization scheme and investigate methods for frequency selection in reduced-order FDMs. Our results noticeably streamline the design process of FDMs, pruning redundant transforms, and leading to speedups of 3x to 10x that increase with data resolution and model size. We perform extensive experiments on learning the solution operator of spatio-temporal dynamics, including incompressible Navier-Stokes, turbulent flows around airfoils and high-resolution video of smoke. T1 models improve on the test performance of FDMs while requiring significantly less computation (5 hours instead of 32 for our large-scale experiment), with over 20% reduction in average predictive error across tasks.

cs.LG↗

arXiv2022

Meta-SysId: A Meta-Learning Approach for Simultaneous Identification and Prediction

In this paper, we propose Meta-SysId, a meta-learning approach to model sets of systems that have behavior governed by common but unknown laws and that differentiate themselves by their context. Inspired by classical modeling-and-identification approaches, Meta-SysId learns to represent the common law through shared parameters and relies on online optimization to compute system-specific context. Compared to optimization-based meta-learning methods, the separation between class parameters and context variables reduces the computational burden while allowing batch computations and a simple training scheme. We test Meta-SysId on polynomial regression, time-series prediction, model-based control, and real-world traffic prediction domains, empirically finding it outperforms or is competitive with meta-learning baselines.

cs.LG↗

arXiv2022

Neural Solvers for Fast and Accurate Numerical Optimal Control

Synthesizing optimal controllers for dynamical systems often involves solving optimization problems with hard real-time constraints. These constraints determine the class of numerical methods that can be applied: computationally expensive but accurate numerical routines are replaced by fast and inaccurate methods, trading inference time for solution accuracy. This paper provides techniques to improve the quality of optimized control policies given a fixed computational budget. We achieve the above via a hypersolvers approach, which hybridizes a differential equation solver and a neural network. The performance is evaluated in direct and receding-horizon optimal control tasks in both low and high dimensions, where the proposed approach shows consistent Pareto improvements in solution accuracy and control performance.

math.OC↗

arXiv2022

Design of a Littrow-type diode laser with independent control of cavity length and grating rotation

We present a novel extended-cavity diode laser (ECDL) based on a modified Littrow configuration. The coarse wavelength adjustment via the rotation of a diffraction grating is decoupled from the fine tuning of the external cavity modes by positioning a piezo transducer behind the diode laser, making the laser robust against misalignment and hysteresis even with long external cavities. Two laser prototypes with external cavities of different lengths were tested with a 780 nm laser diode, and locked to an atomic reference. We observe a mode-hop-free frequency tunability broader than the free spectral range of the external cavity upon changes of its length. The design is well suited to atomic and molecular experiments demanding a high level of stability over time.

physics.ins-det↗

arXiv2021

Prospects for single photon sideband cooling of optically trapped neutral atoms

We propose a novel cooling scheme for realising single photon sideband cooling on particles trapped in a state-dependent optical potential. We develop a master rate equation from an ab-initio model and find that in experimentally feasible conditions it is possible to drastically reduce the average occupation number of the vibrational levels by applying a frequency sweep on the cooling laser that sequentially cools all the motional states. Notably, this cooling scheme works also when a particle experiences a deeper trap in its internal ground state than in its excited state, a condition for which conventional single photon sideband cooling does not work. In our analysis, we consider two cases: a two-level particle confined in an optical tweezer and Li atoms confined in an optical lattice, and find conditions for efficient cooling in both cases. The results from the model are confirmed by a full quantum Monte Carlo simulation of the system Hamiltonian. Our findings provide an alternative cooling scheme that can be applied in principle to any particle, e.g. atoms, molecules or ions, confined in a state-dependent optical potential.

cond-mat.quant-gas↗

arXiv2019

Prospects for single photon sideband cooling in fermionic Lithium

We present an analytic and numerical study for realizing single photon sideband cooling in an ultracold sample of fermionic Lithium trapped in a periodic optical potential. We develop an analytical model and obtain a master equation for the bound level populations. The cooling sequence is simulated both with a laser at a fixed frequency and with a frequency sweep. Finally, a Monte Carlo simulation is performed taking into account the full hyperfine spectrum of $^6\!$Li. We find that a gas of $^6\!$Li atoms loaded from a Magneto-Optical trap into a deep optical lattice can be cooled down to a $99\%$ occupancy of the lattice ground state after a 5mm single photon sideband cooling using the D1 line of Li.

physics.atom-ph↗