arXiv subjects

Rares Dolga

Publications and source records attributed to Rares Dolga.

8 recordsLinked to original sources

arXiv2026

RotRNN: Modelling Long Sequences with Rotations

Linear recurrent neural networks, such as State Space Models (SSMs) and Linear Recurrent Units (LRUs), have recently shown state-of-the-art performance on long sequence modelling benchmarks. Despite their success, their empirical performance is not well understood and they come with a number of drawbacks, most notably their complex initialisation and normalisation schemes. In this work, we address some of these issues by proposing RotRNN -- a linear recurrent model which utilises the convenient properties of rotation matrices. We show that RotRNN provides a simple and efficient model with a robust normalisation procedure, and a practical implementation that remains faithful to its theoretical derivation. RotRNN also achieves competitive performance to state-of-the-art linear recurrent models on several long sequence modelling datasets.

cs.LG↗

arXiv2026

wd1: Weighted Policy Optimization for Reasoning in Diffusion Language Models

Improving the reasoning capabilities of diffusion-based large language models (dLLMs) through reinforcement learning (RL) remains an open problem. The intractability of dLLMs likelihood function necessitates approximating the current, old, and reference policy likelihoods at each policy optimization step. This reliance introduces additional computational overhead, and can lead to large variance and estimation error in RL objective -- particularly in computing the policy ratio for importance sampling. To mitigate these issues, we introduce wd1, a novel ratio-free policy optimization approach that reformulates the RL objective as a weighted log-likelihood, requiring only a single approximation for the current parametrized policy likelihood. We formally show that our proposed method can be interpreted as energy-guided discrete diffusion training combined with negative sample unlearning, thereby confirming its theoretical soundness. In experiments on LLaDA-8B model, wd1 outperforms diffusion-based GRPO (d1) while requiring lower computational cost, achieving up to a $+59\%$ improvement in accuracy. Furthermore, we extend wd1 to denoising-stepwise weighted policy optimization (wd1++), achieving state-of-the-art math performance of $44.2\%$ on MATH500 and $84.5\%$ on GSM8K with only 20 RL training steps.

cs.LG↗

arXiv2025

Unifying Linear-Time Attention via Latent Probabilistic Modelling

Transformers have achieved state-of-the-art results across a range of domains, but their quadratic attention mechanism poses significant challenges for long-sequence modelling. Recent efforts to design linear-time attention mechanisms have yielded more scalable alternatives, yet often at the cost of performance, particularly on discrete data such as language. In this work, we revisit linear attention through the lens of probabilistic graphical models. We first show that standard linear attention can be interpreted as an undirected latent variable model, revealing a key limitation: the absence of directionality. To address this, we propose a novel directed parameterisation of linear attention that introduces an asymmetric structure, enabling an interpretation aligned with the causal and sequential nature of language. Our formulation integrates global latent-variable attention with local standard attention in a fully probabilistic framework. Additionally, we introduce a recurrent parameterisation of queries and keys that avoids reliance on relative positional encodings, often incompatible with linear attention. Experiments on language modelling benchmarks demonstrate that our model achieves competitive performance with standard attention and outperforms existing linear attention variants.

cs.CL↗

arXiv2025

Incremental Sequence Classification with Temporal Consistency

We address the problem of incremental sequence classification, where predictions are updated as new elements in the sequence are revealed. Drawing on temporal-difference learning from reinforcement learning, we identify a temporal-consistency condition that successive predictions should satisfy. We leverage this condition to develop a novel loss function for training incremental sequence classifiers. Through a concrete example, we demonstrate that optimizing this loss can offer substantial gains in data efficiency. We apply our method to text classification tasks and show that it improves predictive accuracy over competing approaches on several benchmark datasets. We further evaluate our approach on the task of verifying large language model generations for correctness in grade-school math problems. Our results show that models trained with our method are better able to distinguish promising generations from unpromising ones after observing only a few tokens.

cs.LG↗

arXiv2025

From Characters to Tokens: Dynamic Grouping with Hierarchical BPE

Subword tokenization methods like Byte Pair Encoding (BPE) are widely used in large language models due to their balance of vocabulary compactness and representational power. However, they suffer from inefficiencies in representing rare words and require large embedding matrices. Character-level models address these issues but introduce performance bottlenecks, particularly in Transformer-based architectures. Recent hierarchical models attempt to merge the benefits of both paradigms by grouping characters into patches, but existing patching strategies either rely on whitespace-limiting applicability to certain languages, or require auxiliary models that introduce new dependencies. In this paper, we propose a dynamic character grouping method that leverages the structure of existing BPE tokenization without requiring additional models. By appending explicit end-of-patch markers to BPE tokens and introducing a second-level BPE compression stage to control patch granularity, our method offers efficient, flexible, and language-agnostic representations. Empirical results demonstrate that our approach matches or exceeds the performance of dynamic entropy- and whitespace-based patching strategies, while maintaining a compact vocabulary.

cs.CL↗

arXiv2025

When Embedding Models Meet: Procrustes Bounds and Applications

Embedding models trained separately on similar data often produce representations that encode stable information but are not directly interchangeable. This lack of interoperability raises challenges in several practical applications, such as model retraining, partial model upgrades, and multimodal search. Driven by these challenges, we study when two sets of embeddings can be aligned by an orthogonal transformation. We show that if pairwise dot products are approximately preserved, then there exists an isometry that closely aligns the two sets, and we provide a tight bound on the alignment error. This insight yields a simple alignment recipe, Procrustes post-processing, that makes two embedding models interoperable while preserving the geometry of each embedding space. Empirically, we demonstrate its effectiveness in three applications: maintaining compatibility across retrainings, combining different models for text retrieval, and improving mixed-modality search, where it achieves state-of-the-art performance.

cs.LG↗

arXiv2024

Code Revert Prediction with Graph Neural Networks: A Case Study at J.P. Morgan Chase

Code revert prediction, a specialized form of software defect detection, aims to forecast or predict the likelihood of code changes being reverted or rolled back in software development. This task is very important in practice because by identifying code changes that are more prone to being reverted, developers and project managers can proactively take measures to prevent issues, improve code quality, and optimize development processes. However, compared to code defect detection, code revert prediction has been rarely studied in previous research. Additionally, many previous methods for code defect detection relied on independent features but ignored relationships between code scripts. Moreover, new challenges are introduced due to constraints in an industry setting such as company regulation, limited features and large-scale codebase. To overcome these limitations, this paper presents a systematic empirical study for code revert prediction that integrates the code import graph with code features. Different strategies to address anomalies and data imbalance have been implemented including graph neural networks with imbalance classification and anomaly detection. We conduct the experiments on real-world code commit data within J.P. Morgan Chase which is extremely imbalanced in order to make a comprehensive comparison of these different approaches for the code revert prediction problem.

cs.SE↗

arXiv2024

Log Summarisation for Defect Evolution Analysis

Log analysis and monitoring are essential aspects in software maintenance and identifying defects. In particular, the temporal nature and vast size of log data leads to an interesting and important research question: How can logs be summarised and monitored over time? While this has been a fundamental topic of research in the software engineering community, work has typically focused on heuristic-, syntax-, or static-based methods. In this work, we suggest an online semantic-based clustering approach to error logs that dynamically updates the log clusters to enable monitoring code error life-cycles. We also introduce a novel metric to evaluate the performance of temporal log clusters. We test our system and evaluation metric with an industrial dataset and find that our solution outperforms similar systems. We hope that our work encourages further temporal exploration in defect datasets.

cs.SE↗