搜索结果：farewell

Recent studies on scaling up ranking models have achieved substantial improvement for recommendation systems and search engines. However, most large-scale ranking systems rely on item IDs, where each item is treated as an independent categorical symbol and mapped to a learned embedding. As items rapidly appear and disappear, these embeddings become difficult to train and maintain. This instability impedes effective learning of neural network parameters and limits the scalability of ranking models. In this paper, we show that semantic tokens possess greater scaling potential compared to item IDs. Our proposed framework TRM improves the token generation and application pipeline, leading to 33% reduction in sparse storage while achieving 0.85% AUC increase. Extensive experiments further show that TRM could consistently outperform state-of-the-art models when model capacity scales. Finally, TRM has been successfully deployed on large-scale personalized search engines, yielding 0.26% and 0.75% improvement on user active days and change query ratio respectively through A/B test.

The dual wave-particle nature of quantum objects is a notoriously unintuitive feature of quantum theories. However, it is often deemed essential, due to quantum objects exhibiting diffraction and interference. We extend the work of Landé and Lévy-Leblond to demonstrate that de Broglie wavelengths are not relativistically covariant as simultaneous spatial structures, making wave properties an unviable explanation of apparent interference. We then explore whether modern experiments vindicate an alternative view: that apparent waviness in diffraction and interference scenarios emerges as a consequence of quantised interactions between particles. Such a view has historically received very little attention, despite being the exact modern explanation of both the Kapitza-Dirac effect and ultrafast electron diffraction. We then study a photon orbital angular momentum realisation of the double slit to show that quantised exchanges can mimic interference. Finally, we demonstrate that the quantum formalism demands that particle momentum is determined at the point of scattering, contravening wave-based explanations of quantum interference.

Session-based recommendation (SBR) aims to predict anonymous users' next interaction based on their interaction sessions. In the practical recommendation scenario, low-exposure items constitute the majority of interactions, creating a long-tail distribution that severely compromises recommendation diversity. Existing approaches attempt to address this issue by promoting tail items but incur accuracy degradation, exhibiting a "see-saw" effect between long-tail and accuracy performance. We attribute such conflict to session-irrelevant noise within the tail items, which existing long-tail approaches fail to identify and constrain effectively. To resolve this fundamental conflict, we propose \textbf{HID} (\textbf{H}ybrid \textbf{I}ntent-based \textbf{D}ual Constraint Framework), a plug-and-play framework that transforms the conventional "see-saw" into "win-win" through introducing the hybrid intent-based dual constraints for both long-tail and accuracy. Two key innovations are incorporated in this framework: (i) \textit{Hybrid Intent Learning}, where we reformulate the intent extraction strategies by employing attribute-aware spectral clustering to reconstruct the item-to-intent mappin

We argue that the principal application for blockchain technology will not be in the financial sector, but rather in maintaining decentralized human governance, from archives to transparent policies encoded in the blockchain in the form of smart contracts.. Such decentralized, blockchain-grounded governance comes not a moment too soon, as nation states are dissolving before our eyes. Will blockchain-based communities replace the nation state? What are the prospects and dangers of this development?

Over the past decades, the volume-of-fluid (VOF) method has been the method of choice for simulating atomization processes, owing to its unique ability to discretely conserve mass. Current state-of-the-art VOF methods, however, rely on the piecewise-linear interface calculation (PLIC) to represent the interface used when calculating advection fluxes. This renders the estimated curvature of the transported interface zeroth-order accurate at best, adversely impacting the simulation of surface-tension-driven flows. In the past few years, there have been several attempts at using piecewise-parabolic interface approximations instead of piecewise-linear ones for computing advection fluxes, albeit all limited to two-dimensional cases or not inherently mass conservative. In this contribution, we present our most recent work on three-dimensional piecewise-parabolic interface reconstruction and apply it in the context of the VOF method. As a result of increasing the order of the interface representation, the reconstruction of the interface and the estimation of its curvature now become a single step instead of two separate ones. The performance of this new approach is assessed both in terms

We reformulate our proof of the under-determination of Cotton gravity in terms of the Codazzi parametrization.

Superstrata microstate geometries furnish some of the most successful laboratories, to date, for probing black hole microstructure in a geometric setting. This paper extends the (1,m,n) family of superstrata, to allow for flat asymptotics. Previous constructions utilized the decoupling regime, where the geometry is asymptotically AdS3xS3. Brief comments are made on the additional complexity introduced by the flat space coupling, how this obscures holomorphicity and breaks the consistent truncation to which the decoupled solutions belong. Holomorphicity and consistent truncation were key simplifications assisting previous studies of the (1,m,n) superstrata, undertaken in the decoupling regime. Further, these results open a window for future projects to determine how previous analysis on the decoupled geometries extend or are modified once flat space asymptotics are imposed. Since our universe is almost flat on cosmological scales, this represents progress towards more phenomenologically relevant microstate geometries. This work can be considered a continuation of that in arxiv.org/abs/1711.10474 where some single mode superstrata were also coupled to flat space.

We outline the course of affairs in the experimental and theoretical fields of nuclear and particle physics which determined its finale, and give several fragmentary remarks on its present state. The essay tells about events and their participants, known from the literature, but presented here from the perspective of a person whose 50-year labor activity, 1972--2022, proceeded at the All-Union Scientific Research Institute of Experimental Physics, Sarov. In order to predict the fate of particle physics and related astrophysics and cosmology, it is useful to become aware of the facts about another branch of physics that has already gone through its decline $\sim 30$ years ago, the physics of nuclear weapons. These facts are important not in themselves, but as evidence of the growing problems of science and social life, which are not only far from a satisfactory solution, but have not even been the subject of any serious discussion.

Pretrained language models have achieved remarkable success in various natural language processing tasks. However, pretraining has recently shifted toward larger models and larger data, and this has resulted in significant computational and energy costs. In this paper, we propose Influence Subset Selection (ISS) for language model, which explicitly utilizes end-task knowledge to select a tiny subset of the pretraining corpus. Specifically, the ISS selects the samples that will provide the most positive influence on the performance of the end-task. Furthermore, we design a gradient matching based influence estimation method, which can drastically reduce the computation time of influence. With only 0.45% of the data and a three-orders-of-magnitude lower computational cost, ISS outperformed pretrained models (e.g., RoBERTa) on eight datasets covering four domains.

We discuss a dramatic change brought into the pQCD description of hard processes in a nuclear environment by a large thickness of heavy nuclei. It breaks the familiar linear $k_{\perp}$-factorization which must be replaced by a new concept of the nonlinear $k_{\perp}$-factorization introduced in \cite{Nonlinear}.We demonstrate the salient features of nonlinear $k_{\perp}$-factorization on several examples from hard dijet production in DIS off heavy nuclei to single-jet to dijet production in hadron-nucleus collisions >. We also comment briefly on the non-linear BFKL evolution for gluon density of nuclei.

The algebraic formulation of Wick's theorem that allows one to present the vacuum or thermal averages of the chronological product of an arbitrary number of field operators as a determinant (permanent) of the matrix is proposed. Each element of the matrix is the average of the chronological product of only two operators. This formulation is extremely convenient for practical calculations in quantum field theory and statistical physics by the methods of symbolic mathematics using computers.