arXiv:2604.16883v1 Announce Type: new
Abstract: In long-context decoding for LLMs and LMMs, attention becomes increasingly memory-bound because each decoding step must load a large amount of KV-cache data from GPU memory. Existing acceleration strategies often trade efficiency for accuracy by relying on heuristic pruning that may discard useful information. At a deeper level, they also tend to indiscriminately preserve all high-scoring tokens, treat early tokens as indispensable anchors, or ...
arXiv:2604.16881v1 Announce Type: new
Abstract: Cross-cultural entity translation remains challenging for large language models (LLMs) as literal or phonetic renderings are usually yielded instead of culturally appropriate translations in context. However, relevant knowledge may already be encoded in model parameters during large-scale pre-training. To incentivize the effective use of parametric knowledge, we propose EA-RLVR (Entity-Anchored Reinforcement Learning with Verifiable Rewards), a...
arXiv:2604.16880v1 Announce Type: new
Abstract: Ring-based collective operations are widely used in distributed AI training due to their efficient bandwidth utilization. While ring communication excels at pipelining, its performance is heavily dependent on having synchronized step-wise progression. This presents a mismatch to the underlying network conditions in practice: collective operations are vulnerable to network jitter and congestion, leading to step misalignment and increased collect...
arXiv:2604.16879v1 Announce Type: new
Abstract: With the rapid development of generative models and multimodal content editing technologies, the key challenge faced by synthetic image detection (SID) lies in cross-distribution generalization to unknown generation sources. In recent years, visual foundation models (VFM), which acquire rich visual priors through large scale image-text alignment pretraining, have become a promising technical route for improving the generalization ability of SID...
arXiv:2604.16878v1 Announce Type: new
Abstract: Early prediction of severe clinical deterioration and remaining length of stay can enable timely intervention and better resource allocation in high-acuity settings such as the ICU. This has driven the development of machine learning models that leverage continuous streams of vital signs and other physiological signals for real-time risk prediction. Despite their promise, existing methods have important limitations. Contrastive pretraining trea...
arXiv:2604.16875v1 Announce Type: new
Abstract: A central question in computational neuroscience is whether the learning rule used to train a neural network determines how well its internal representations align with those of the human visual cortex. We present a systematic comparison of four learning rules -- backpropagation (BP), feedback alignment (FA), predictive coding (PC), and spike-timing-dependent plasticity (STDP) -- applied to identical convolutional architectures and evaluated ag...
arXiv:2604.16872v1 Announce Type: new
Abstract: Large Language Models (LLMs) can be helpful for literature search and summarisation, but retracted articles can confuse them. This article asks three open weights (offline) LLMs whether 161 high profile retracted articles had been retracted, performing a similar check for a benchmark multidisciplinary set of 34,070 non-retracted articles. Based on titles and abstracts, in over 80% of cases the LLMs claimed that a retracted article had not been ...
arXiv:2604.16871v1 Announce Type: new
Abstract: Neuro-symbolic Reinforcement Learning (NeSy-RL) combines symbolic reasoning with gradient-based optimization to achieve interpretable and generalizable policies. Relational concepts, such as "left of" or "close by", serve as foundational building blocks that structure how agents perceive and act. However, conventional approaches require human experts to manually define these concepts, limiting adaptability since concept semantics vary across en...
pg_plan_advice is expected to land in PostgreSQL 19. That makes this a good moment to look at query hints — what they are, what every other major database does with them, and how PostgreSQL ended up being the obvious outlier. Three parts. This is the first.
What a hint is
A query hint is an instr…
For most of PostgreSQL’s history, the official community position on query hints has been a polite version of “no, and stop asking.”
The position isn’t subtle. The PostgreSQL wiki maintains a page titled Not Worth Doing, and “Oracle-style optimizer hints” is li…
We begin at allow_alter_system, which is both new and politically fraught — so let’s start with a fight.
ALTER SYSTEM was added in 9.4 as a quality-of-life improvement: set GUCs from an SQL prompt, have the values written into postgresql.auto.conf, no shell access required. It was immediate…
pg_plan_advice is expected to land in PostgreSQL 19. That makes this a good moment to look at query hints — what they are, what every other major database does with them, and how PostgreSQL ended up being the obvious outlier. Three parts. This is the first.
What a hint is
A query hint is an instr…
Questions about methodologies and findings in computing and other fields have pushed reproducibility from once-obscure corners of academia into the public spotlight.
On a PostgreSQL build with assertions enabled, run the standard make check-world suite with a small debugging extension called pg_pathcheck loaded. It will report on pointers to freed memory in the planner's path lists. Such dangling pointers exist even in the core Postgres now. They are harmless today. But the word today is what makes this worth writing about.A production storyThis story started in July 2021. At the time, I was finishing a sharding solution built on top of postgres_fdw. Dur...
Everyone who’s gone at least knee-deep in photography knows there’s this idea of the exposure triangle: aperture, shutter speed, and ISO. Depending on what you’re going for artistically, you adjust the three parameters, knowing that there are trade-offs in doing so. After working on a few cases, and presenting solutions to customers, I’ve started to think about Postgres performance tuning in a similar way – there are basic parameters that can be tuned, and there are trade-offs for...
Property Based Testing and fuzzing are a deep and science-intensive topic. There are enough advanced techniques there for a couple of PhDs, a PBT daemon, and a client-server architecture. But I have this weird parlor-trick PBT library, implementable in a couple of hundred lines of code in one sitting.
It's been one of those months, and by that, I mean one of the 663 months since I was born. This won't be a long post, because I only have two things to say. First, I'm really glad we re-ordered the GMI (Guaranteed Minimum
In another sign that the depravity economy has no bottom, Forbes published a story about a Louisiana man that killed 8 children over the weekend containing a box that asked readers to predict whether Congress would do anything about gun control. Citation Needed author Molly White first spotted the box
Shanhai Liao | Senior Software Engineer, Content Acquisition and Media Platform; Di Ruan, | Senior Staff Software Engineer, Content Acquisition and Media Platform; Evan Li, | Senior Engineering Manager, Content Acquisition and Media PlatformIntroductionAccurate content understanding underpins Pinterest’s ability to drive distribution and engagement. This requires deep insight not just into the image itself, but also the outbound links or items to which those images point. At the foundation...
AI Inference
