This paper explores learnings from first-generation Compute Express Link (CXL) memory expansion to accelerate CXL’s journey to broad, robust use. While broad adoption will be a long journey similar to that of RDMA, we argue that the first step—CXL.mem expansion—is viable on today’s hardware. Through an end-to-end analysis, we revisit common showstoppers: we decompose memory access latency and show that CPU and DRAM internals, rather than the CXL protocol, dominate latency and variability, and we demonstrate how system slack absorbs link error rates above nominal specifications. Along the way, we distill practical guidance on device validation, monitoring, failure modes, security, and multi-tenant interference, and we outline a pragmatic adoption pathway: solidify robust expansion first, prototype micro-pooling next, and move to selective sharing as the ecosystem matures.

Costas Mavromatis, Soji Adeshina, Vassilis N. Ioannidis, Zhen Han, Qi Zhu, Ian Robinson, Bryan Thompson, Huzefa Rangwala, George Karypis. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Retrieval-Augmented Generation (RAG) grounds large language models in external evidence, yet it still falters when answers must be pieced together across semantically distant documents. We close this gap with the Hierarchical Lexical Graph (HLG), a three-tier index that (i) traces every atomic proposition to its source(ii) clusters propositions into latent topics, and (iii) links entities and relations to expose cross-document paths. On top of HLG we build two complementary, plug-and-play retrievers: StatementGraphRAG, which performs fine-grained entity-aware beam search over propositions for high-precision factoid questions, and TopicGraphRAG, which selects coarse topics before expanding along entity links to supply broad yet relevant context for exploratory queries. Additionally, existing benchmarks lack the complexity required to rigorously evaluate multi-hop summarization systems, often focusing on single-document queries or limited datasets. To address this, we introduce a synthetic dataset generation pipeline that curates realistic, multi-document question-answer pairs, enabling robust evaluation of multi-hop retrieval systems. Extensive experiments across five datasets demonstrate that our methods outperform naive chunk-based RAG, achieving an average relative improvement of 23.1% in retrieval recall and correctness. Open-source Python library is available at https://github.com/awslabs/graphrag-toolkit.

A three-dimensional automatic cubature routine, calledelrint3d, is described and numerical results are presented that demonstrate its applicability across a wide range of domains and integrand types. The underlying algorithm is based on a 2s-copy lattice augmentation sequence, the seed lattice for which has been determined by exhaustive search based on optimization of index of merit and trigonometric degree of precision.

Satellite measurements of global sea surface temperature (SST) distribution are increasingly recognised to have great importance for understanding changes in the world's climate, as well as for operational forecasting of the oceans and atmosphere. If satellite-derived SSTs are to provide the basis of an essential climate variable (ECV) it is necessary to establish methods for independently verifying their quality. Previously the validation of satellite SST products has been performed by comparison with sea temperatures measured by contact thermometers on the hulls of moored or drifting buoys or of ships. Such sensors operate below the sea surface, within the top few metres of the water column, recording what is now referred to [1] as SSTdepth. Infrared radiometers on satellites measure the temperature of the surface skin of the ocean, referred to as SSTskin. However, significant uncertainties are introduced when validation is based on comparisons between SSTdepthand SSTskinbecause the sea temperature is not uniform across the skin and the upper few metres, especially but not only because of diurnal variability. Such uncertainties can be eliminated by using in situ SSTskinmeasurements as the basis for a satellite SST validation approach which genuinely compares like with like. This paper describes the principles and results of such a system which has now been in place for five years. It has been used to validate SST measurements from the Advanced Along Track Scanning Radiometer (AATSR) on Envisat. Within the collaborative approach developed by the Group for High Resolution SST (GHRSST, [1]) for merging of complementary data from different satellites, the primary role of AATSR data is to provide a reference for bias adjustment of other data. Precise validation of AATSR measurements is therefore essential for maintaining the quality of other GHRSST products.

The Advanced Along Track Scanning Radiometer is an imaging radiometer on board ESA's Envisat satellite. The primary mission aim for AATSR is the measurement of sea surface temperature (SST), on a global scale, to the degrees of accuracy and precision required for climate change research (better than 0.3K ± 1 sigma). Since the launch of Envisat, on 1st March 2002, AATSR has been producing excellent data, with images of high quality in the thermal, near infrared and visible wavelength regions. An intensive validation programme has been underway to assess whether AATSR SST data meet their stringent accuracy requirements. There have been a number of activities for SST validation using, for example, SST data from buoys, global SST analysis fields and SST measurements from high-precision radiometers on research vessels and ships of opportunity. Initial assessment indicates that the AATSR SST shows a global mean deviation of better than 0.3K with respect to in situ observations, with significantly higher levels of accuracy under certain conditions, thereby meeting its performance objectives.

Abstract Although it has long been realized that ACID (atomicity, consistency, isolation, durability) transactions by themselves are not adequate for structuring long‐lived applications and much research work has been done on developing specific extended transaction models, no middleware support for building extended transactions is currently available and the situation remains that a programmer often has to develop application specific mechanisms. The CORBA Activity Service Framework described in this paper is a way out of this situation. The design of the service is based on the insight that the various extended transaction models can be supported by providing a general purpose event signalling mechanism that can be programmed to enable activities—application specific units of computations—to coordinate each other in a manner prescribed by the model under consideration. The different extended transaction models can be mapped onto specific implementations of this framework, permitting such transactions to span a network of systems connected indirectly by some distribution infrastructure. The framework described in this paper is an overview of the OMG's (Object Management Group) Additional Structuring Mechanisms for the OTS standard. Through a number of examples the paper shows that the framework has the flexibility to support a wide variety of extended transaction models. Although the framework is presented here in CORBA specific terms, the main ideas are sufficiently general, so that it should be possible to use them in conjunction with other middleware. Copyright © 2003 John Wiley & Sons, Ltd.

r2d2lriis a non-adaptive algorithm implemented in C++ for performing automatic cubature over a wide variety of finite and non-finite two-dimensional domains. The core integrator uses a sixth-order Sidi transformation applied to a sequence of embedded lattice rules in such a fashion as to incur virtually no computational overhead. Even for integrals over non-finite domains, for which several non-finite to finite transformations may be attempted, the algorithm remains very fast. Performance data are presented which demonstrate both the effectiveness and efficiency ofr2d2lri, taking into account the number of function evaluations needed and the execution speed.

as T R I E X ( ) for a subroutine or a function, R E S U L T for a variable name, and V E R ( , , ,) for a two-dimensional array.The integrator T R I E X ( ) is coded as a F O R T R A N subroutine using double precision arithmetic.

We describe an automatm algorithm for integration over a triangle.The algorithm uses an adaptive subdwlsional strategy with global acceptance criteria and incorporates the e-algorithm to speed convergence.Numerical results are presented which show that it deals effectively with various types of singularity along the boundaries of the integration triangle (or along the boundaries of eventual subtriangles of the subdivision).