arXiv subjects

Mateusz Kowalczyk

Publications and source records attributed to Mateusz Kowalczyk.

6 recordsLinked to original sources

arXiv2026

SAGE-32B: Agentic Reasoning via Iterative Distillation

We demonstrate SAGE-32B, a 32 billion parameter language model that focuses on agentic reasoning and long range planning tasks. Unlike chat models that aim for general conversation fluency, SAGE-32B is designed to operate in an agentic loop, emphasizing task decomposition, tool usage, and error recovery. The model is initialized from the Qwen2.5-32B pretrained model and fine tuned using Iterative Distillation, a two stage training process that improves reasoning performance through rigorously tested feedback loops. SAGE-32B also introduces an inverse reasoning approach, which uses a meta cognition head to forecast potential failures in the planning process before execution. On agentic reasoning benchmarks including MMLU-Pro, AgentBench, and MATH-500, SAGE-32B achieves higher success rates in multi tool usage scenarios compared to similarly sized baseline models, while remaining competitive on standard reasoning evaluations. Model weights are publicly released at https://huggingface.co/sagea-ai/sage-reasoning-32b

cs.AI↗

arXiv2026

SAGE Celer 2.6 Technical Card

We introduce SAGE Celer 2.6, the latest in our line of general-purpose Celer models from SAGEA. Celer 2.6 is available in 5B, 10B, and 27B parameter sizes and benefits from extensive architectural modifications and further pre-training on an undisclosed model. Using our Inverse Reasoning (IR) pipeline, SAGEA natively trains Celer 2.6 to validate its own logic paths, minimizing cascading error and hallucination in complex reasoning tasks. Celer 2.6 also boasts natively integrated multimodal functionality with an end-to-end vision encoder to avoid common pitfalls in adapter-based approaches. Celer 2.6 provides highly competitive results on mathematics, coding, and general intelligence benchmarks (ACUMEN), along with low latency. Most importantly, Celer 2.6 is specifically optimized for South Asian language support, with a custom tokenizer for the Devanagari script and strong performance in both Nepali and Hindi without sacrificing English reasoning ability.

cs.CL↗

arXiv2025

How Likely Are You to Observe Non-locality with Imperfect Detection Efficiency and Random Measurement Settings?

Imperfect detection efficiency remains one of the major obstacles in achieving loophole-free Bell tests over long distances. At the same time, the challenge of establishing a common reference frame for measurements becomes more pronounced as the separation between parties increases. In this work, we tackle both of these issues by examining the impact of limited detection efficiency on the probability of Bell inequality violation with Haar random measurement settings. We derive analytical lower bounds on the violation probability for a two-qubit maximally entangled state, which is tight for correlation inequalities and perfect detection efficiencies. We further investigate it numerically for more qubits and settings using two detection efficiency models and an original method based on linear programming. Beyond that, we show that the so-called typicality of Bell inequality violation, i.e., almost certain violation of local realism with sufficiently many particles or random measurement directions, holds even if the detection efficiency is limited. Our findings reveal that increasing the number of measurement settings can compensate for efficiency losses above the critical threshold. We determine critical detection efficiencies for both two-party and three-party scenarios. For two parties, we recover previously established results, while for three parties, we derive a symmetric critical efficiency of $η_{crit} = 2/3$ for the W and GHZ states within the binning model. In cases involving the no-detection outcome, we present a modified inequality using a pair of orthogonal observables for each party with $η_{crit} \approx 0.7208$, which is notably less than 0.75 for the GHZ state. These results offer deeper insights into the limitations and possibilities for certifying non-locality in the presence of limited detection efficiency, shedding light on its robustness in practical Bell tests.

quant-ph↗

arXiv2024

Unmasking the Polygamous Nature of Quantum Nonlocality

Quantum mechanics imposes limits on the statistics of certain observables. Perhaps the most famous example is the uncertainty principle. Similar trade-offs also exist for the simultaneous violation of multiple Bell inequalities. In the simplest case of three observers, it has been shown that if two observers violate a Bell inequality then none of them can violate any Bell inequality with the third observer, a property called monogamy of Bell violations. Forms of Bell monogamy have been linked to the no-signalling principle and the inability of simultaneous violations of all inequalities is regarded as their fundamental property. Here we show that the Bell monogamy does not hold universally and that in fact the only monogamous situation exists for only three observers. Consequently, the nature of quantum nonlocality is truly polygamous. We present a systematic methodology for identifying quantum states, measurements and tight Bell inequalities that do not obey the monogamy principle for any number of more than three observers. The identified polygamous inequalities enable any subset of $(N-1)$ observers to reveal nonlocality, which is also shown experimentally by measuring Bell-type correlations of six-photon Dicke states. Our findings may be exploited for multiparty quantum key distribution as well as simultaneous self-testing of multiple nodes in quantum networks.

quant-ph↗

arXiv2024

SoccerNet 2024 Challenges Results

The SoccerNet 2024 challenges represent the fourth annual video understanding challenges organized by the SoccerNet team. These challenges aim to advance research across multiple themes in football, including broadcast video understanding, field understanding, and player understanding. This year, the challenges encompass four vision-based tasks. (1) Ball Action Spotting, focusing on precisely localizing when and which soccer actions related to the ball occur, (2) Dense Video Captioning, focusing on describing the broadcast with natural language and anchored timestamps, (3) Multi-View Foul Recognition, a novel task focusing on analyzing multiple viewpoints of a potential foul incident to classify whether a foul occurred and assess its severity, (4) Game State Reconstruction, another novel task focusing on reconstructing the game state from broadcast videos onto a 2D top-view map of the field. Detailed information about the tasks, challenges, and leaderboards can be found at https://www.soccer-net.org, with baselines and development kits available at https://github.com/SoccerNet.

cs.CV↗

arXiv2016

A new fast algorithm for reproducing complex networks with community structure

In this paper, we introduce a new algorithm allowing for generation of networks with heterogeneity of both node degrees and community sizes. The quality and efficiency of the algorithm is analyzed and compared to the other, so far the most popular algorithm which was proposed by Lancichinetti et al. We discuss the advantages and shortcomings of both algorithms indicating the areas of their potential application.

physics.soc-ph↗