Search for: All records

Speculative Decoding (SD) enforces strict distributional equivalence to the target model when accepting candidate tokens. While it maintains the target model’s generation quality, this strict equivalence limits the speedup achievable by SD and prevents users from trading deviations from the target distribution in exchange for further inference speed gains. To address these limitations, we introduce Fuzzy Speculative Decoding (FSD) - a decoding algorithm that generalizes SD by accepting candidate tokens based on the divergences between the target and draft model distributions. By allowing for controlled divergence from the target model, FSD enables users to flexibly trade generation quality for inference speed. Across several benchmarks, our method is able to achieve significant runtime improvements of over 5 tokens per second faster than SD at only an approximate 2% absolute reduction in benchmark accuracy. In many cases, FSD is even able to match SD benchmark accuracy at over 2 tokens per second faster, demonstrating that distributional equivalence is not necessary to maintain target model performance. Furthermore, FSD can be seamlessly integrated into existing SD extensions; we demonstrate this by applying FSD to EAGLE-2, greatly enhancing this existing extension’s efficiency while allowing it to leverage FSD’s tunable quality-speed trade-off. 

Synthesis of Ag–Au nanoparticles of different sizes, compositions, element configurations (alloy- and core–shell-like) and electrocatalytic activities by using precursor-loaded PS–P2VP reverse micelles as nanoreactors. 

In everyday life, people routinely make decisions that involve irredeemable risks such as death (e.g., while driving). Even though these decisions under extinction risk are common, practically important, and have different properties compared to the types of decisions typically studied by decision scientists, they have received little research attention. The present work advances the formal understanding of decision making under extinction risk by introducing a novel experimental paradigm, the Extinction Gambling Task (EGT). We derive optimal strategies for three different types of extinction and near-extinction events, and compare them to participants’ choices in three experiments. Leveraging computational modelling to describe strategies at the individual level, we document strengths and shortcomings in participants’ decisions under extinction risk. Specifically, we find that, while participants are relatively good in terms of the qualitative strategies they employ, their decisions are nevertheless affected by loss chasing, scope insensitivity, and opportunity cost neglect. We hope that by formalising decisions under extinction risk and providing a task to study them, this work will facilitate future research on an important topic that has been largely ignored. 

Open clusters (OCs) act as key probes that can be leveraged to constrain the formation and evolution of the Milky Way (MW)’s disk, as each has a unique chemical fingerprint and well-constrained age. Significant Galactic dynamic interactions can leave imprints on the orbital properties of OCs, allowing us to use the present-day properties of long-lived OCs to reconstruct the MW’s dynamic history. To explore these changes, we identify OC analogs in FIRE-2 simulations of MW-mass galaxies. For this work, we focus on one particular FIRE-2 OC, which we identify as an analog to the old, subsolar, distant, and high-Galactic-latitude MW OC, Berkeley 20. Our simulated OC resides ∼6 kpc from the galactic center and ultimately reaches a height $\mid Z_{\max }\mid >2$kpc from the galactic disk, similar to Berkeley 20. We trace the simulated cluster’s orbital and environmental history, identifying key perturbative episodes, including (1) an interaction with a gas overdensity in a spiral arm that prompts an outward migration event and (2) a substantial interaction with a Sagittarius Dwarf Spheroidal Galaxy–mass satellite that causes significant orbital modification. Our simulated OC shows significant resilience to disruption during both its outward migration and the satellite-driven heating event that causes subsequent inward migration. Ultimately, we find these two key processes—migration and satellite heating—are essential to include when assessing OC orbital dynamics in the era of Gaia. 

We survey the opportunities offered by the detection of the forward muons that accompany the creation of neutral effective vector bosons at a muon collider, in different kinematic regimes. Vectors with relatively low energy produce the Higgs boson and the extended muon angular coverage enables studies of the Higgs properties, such as the measurement of the inclusive production cross section and the branching ratio to invisible final states. New heavy particles could be produced by vectors of higher energy, through Higgs portal interactions. If the new particles are invisible, the detection of the forward muons is essential in order to search for this scenario. The angular correlations of the forward muons are sensitive to the quantum interference between the vector-boson helicity amplitudes and can be exploited for the characterization of vector-boson scattering and fusion processes. This is illustrated by analyzing the $CP$properties of the Higgs coupling to the $Z$boson. Our findings provide a physics case and a set of benchmarks for the design of a dedicated forward muon detector. Published by the American Physical Society2025 

Abstract Particle collisions at accelerators like the Large Hadron Collider (LHC), recorded by experiments such as ATLAS and CMS, enable precise standard model measurements and searches for new phenomena. Simulating these collisions significantly influences experiment design and analysis but incurs immense computational costs, projected at millions of CPU-years annually during the high luminosity LHC (HL-LHC) phase. Currently, simulating a single event with Geant4 consumes around 1000 CPU seconds, with calorimeter simulations especially demanding. To address this, we propose a conditioned quantum-assisted generative model, integrating a conditioned variational autoencoder (VAE) and a conditioned restricted Boltzmann machine (RBM). Our RBM architecture is tailored for D-Wave’s Pegasus-structured advantage quantum annealer for sampling, leveraging the flux bias for conditioning. This approach combines classical RBMs as universal approximators for discrete distributions with quantum annealing’s speed and scalability. We also introduce an adaptive method for efficiently estimating effective inverse temperature, and validate our framework on Dataset 2 of CaloChallenge.