Results 41 to 50 of about 2,749 (228)

SPSA View on the Straight-Through Estimator in Neural Network Quantization

IEEE Access
Low-bit quantization-aware training of neural networks typically relies on the straight-through estimator (STE) to learn both quantized weights and their associated scales or effective bit-widths.
Sergey Salishev, Anton Makarov, Elizaveta Tarasova, Kirill Shishin, Oleg Granichin   +4 more
doaj   +1 more source

From tumor‐centric to ecosystem‐based hypotheses in brain tumor research and care

Molecular Oncology, EarlyView.
Primary brain tumors, whether in adults or children, present a major challenge because of their dramatic prognosis and the ongoing lack of efficient therapeutic approaches. In recent years, a shift has occurred from tumor‐centric concepts to a more holistic view of these tumors as dynamic ecosystems.
Julie Gavard, Flavie Bénard, Eva Billat, Margaux Le Guyon, Victoria Maltret, Laura Merlet, Chloé Astrid Walter, Tony Avril, Eric Chevet   +8 more
wiley   +1 more source

Theoretical Time Evolution of Numerical Errors When Using Floating Point Numbers in Shallow‐Water Models

Journal of Advances in Modeling Earth Systems, 2019
We carried out a theoretical investigation of the impact of the numerical errors caused by using floating point numbers (FPNs) in simulations, such as rounding errors.
Tsuyoshi Yamaura, Seiya Nishizawa, Hirofumi Tomita   +2 more
doaj   +1 more source

A stochastic-programming approach for scheduling catch-up rounds in round-robin sport leagues

IMA Journal of Management Mathematics, 2020
AbstractIn most sport leagues, a schedule is announced before the start of the season. However, due to unexpected events (e.g. bad weather conditions), some games cannot be played on the announced date. To handle this, before the start of the season, empty so-called catch-up rounds are positioned in the schedule as a buffer.
Yi, Xiajie, Goossens, Dries
openaire   +2 more sources

Single‐cell DNA methylation profiling: Technologies, computation, and applications in precision oncology

Molecular Oncology, EarlyView.
Single‐cell DNA methylation (scDNAme) profiling maps epimutational clonal evolution, revealing mechanisms of malignancy and therapeutic resistance across diverse cancer types. By providing a high‐resolution landscape of intratumoral heterogeneity, these technologies empower precise patient stratification, guide the development of enhanced ...
Ik Soo Kim
wiley   +1 more source

Understanding the Stochastic Nature of Process Parameter Development of Blown Powder Laser Beam Directed Energy Deposition Additive Manufacturing of Pure Molybdenum

Advanced Engineering Materials, EarlyView.
Identified through the use of statistical design of experiments and metallographic investigation, this study exposes the stochastic origins of intergranular cracks in blown powder laser beam directed energy deposition additive manufacturing of pure molybdenum. It further demonstrates a successful crack mitigation approach with direct correlation to the
Nathaniel J. Lies, Daniel E. Driemeyer, Aaron P. Stebner   +2 more
wiley   +1 more source

Stochastic Multi-round Submodular Optimization with Budget

CoRR
In this work, we study the Stochastic Budgeted Multi-round Submodular Maximization (SBMSm) problem, where we aim to adaptively maximize the sum, over multiple rounds, of a monotone and submodular objective function defined on subsets of items. The objective function also depends on the realization of stochastic events, and the total number of items we ...
Vincenzo Auletta, Diodato Ferraioli, Cosimo Vinci   +2 more
openaire   +2 more sources

Is an Apple an Orange? A Large Language Model Benchmark for Candidate Term Extraction and Subclass Decisions Against Upper Ontologies in Engineering and Materials Science

Advanced Engineering Materials, EarlyView.
Building machine‐readable vocabularies for materials science is slow, expert‐driven work. This study benchmarks 13 large language models on two of its first steps: finding candidate terms in engineering articles and deciding where they belong in a class hierarchy.
Thomas Bjarsch, Maximilian Holland, Klaus Drechsler, Johannes Schilp   +3 more
wiley   +1 more source

Biomimetic Microfractals Based Flexible Triboelectric Nanogenerators

Advanced Functional Materials, EarlyView.
A leaf‐skeleton‐based microfractal architecture is integrated with guided CuNW networks and biomimetic triboelectric polymer layers, enabling breathable, flexible, and material‐efficient triboelectric nanogenerators. The resulting system combines enhanced charge generation and charge transport, offering potential for next‐generation self‐powered ...
Amit Barua, Mislav Matić, Ana‐Marija Pitner, Rituporn Gogoi, Aman Kumar, Mirko Poljak, Anastasia Koivikko, Vipul Sharma   +7 more
wiley   +1 more source

Training LSTM Networks With Resistive Cross-Point Devices

Frontiers in Neuroscience, 2018
In our previous work we have shown that resistive cross point devices, so called resistive processing unit (RPU) devices, can provide significant power and speed benefits when training deep fully connected networks as well as convolutional neural ...
Tayfun Gokmen, Malte J. Rasch, Wilfried Haensch   +2 more
doaj   +1 more source