Note: When clicking on a Digital Object Identifier (DOI) number, you will be taken to an external site maintained by the publisher.
Some full text articles may not yet be available without a charge during the embargo (administrative interval).
What is a DOI Number?
Some links on this page may take you to non-federal websites. Their policies may differ from this site.
-
Do LMs infer the semantics of text from co-occurrence patterns in their training data? Merrill et al. (2022) argue that, in theory, sentence co-occurrence probabilities predicted by an optimal LM should reflect the entailment relationship of the constituent sentences, but it is unclear whether probabilities predicted by neural LMs encode entailment in this way because of strong assumptions made by Merrill et al. (namely, that humans always avoid redundancy). In this work, we investigate whether their theory can be used to decode entailment relations from neural LMs. We find that a test similar to theirs can decode entailment relations between natural sentences, well above random chance, though not perfectly, across many datasets and LMs. This suggests LMs implicitly model aspects of semantics to predict semantic effects on sentence co-occurrence patterns. However, we find the test that predicts entailment in practice works in the opposite direction to the theoretical test. We thus revisit the assumptions underlying the original test, finding its derivation did not adequately account for redundancy in human-written text. We argue that better accounting for redundancy related to explanations might derive the observed flipped test and, more generally, improve computational models of speakers in linguistics.more » « lessFree, publicly-accessible full text available August 11, 2025
-
We propose a method to teach multiple large language models (LLM) to collaborate by interleaving their generations at the token level. We model the decision of which LLM generates the next token as a latent variable. By optimizing the marginal likelihood of a training set under our latent variable model, the base LLM automatically learns when to generate itself and when to call on one of the “assistant” language models to generate, all without direct supervision. Token-level collaboration during decoding allows for a fusion of each model’s expertise in a manner tailored to the specific task at hand. Our collaborative decoding is especially useful in cross-domain settings where a generalist base LLM learns to invoke domain ex- pert models. On instruction-following, domain- specific QA, and reasoning tasks, we show that the performance of the joint system exceeds that of the individual models. Through qualitative analysis of the learned latent decisions, we show models trained with our method exhibit several interesting collaboration patterns, e.g., template-filling.more » « lessFree, publicly-accessible full text available August 11, 2025
-
Abstract As technology becomes increasingly miniaturized, thermal management becomes challenging to keep devices away from overheating due to extremely localized heat dissipation. Two-phase cooling or flow boiling in microspaces utilizes the highly efficient thermal energy transport of phase change from liquid to vapor. However, the excessive consumption of liquid-phase by highly localized heat source causes the two-phase flow maldistribution, leading to a significantly reduced heat transfer coefficient, high-pressure loss, and limited flow rate. In this study, flow boiling in a two-dimensional (2D) microgap heat sink with a hydrophilic coating is investigated with bubble morphology, heat transfer, and pressure drop for conventional (nonhydrophilic) and hydrophilic heat sinks. The experiments are carried out on a stainless steel (SS) plate, having a microgap depth of 170 μm using de-ionized (DI) water at room temperature. Two different hydrophilic surfaces (partial and full channel shape) are fabricated on the heated surface to compare the thermal performance with the conventional surface. Vapor films and slugs are flushed quickly on the hydrophilic surfaces, resulting in heat transfer enhancement on the hydrophilic heat sink compared to the conventional heat sink. The channel hydrophilic heat sink shows better cooling performance and pressure stability as it provides a smooth route for the incoming water to cool the hot spot. Moreover, the artificial neural network (ANN) prediction of heat transfer coefficient shows a good agreement with the experimental results as data fit within ±5% average error.more » « less
-
Quantum information protocols are being deployed in increasingly practical scenarios, via optical fibers or free space, alongside classical communications channels. However, entanglement, the most critical resource to deploy to the communicating parties, is also the most fragile to the noise-induced degradations. Here we show that polarization-frequency hyperentanglement of photons can be effectively employed to enable noise-resistant distribution of polarization entanglement through noisy quantum channels. In particular, we demonstrate that our hyperentanglement-based scheme results in an orders-of-magnitude increase in the SNR for distribution of polarization-entangled qubit pairs, enabling quantum communications even in the presence of strong noise that would otherwise preclude quantum operations due to noise-induced entanglement sudden death. While recent years have witnessed tremendous interest and progress in long-distance quantum communications, previous attempts to deal with the noise have mostly been focused on passive noise suppression in quantum channels. Here, via the use of hyperentangled degrees of freedom, we pave the way toward a universally adoptable strategy to enable entanglement-based quantum communications via strongly noisy quantum channels.