Speculative decoding is widely adopted to reduce latency in large language model (LLM) inference by leveraging smaller draft models capable of handling diverse user tasks. However, emerging AI applications, such as LLM-based agents, present unique workload characteristics: instead of diverse independent requests, agentic frameworks typically submit repetitive inference requests, such as multi-agent pipelines performing similar subtasks or self-refinement loops iteratively enhancing outputs. These workloads result in long and highly predictable sequences, which current speculative decoding methods do not effectively exploit. To address this gap, we introduce SuffixDecoding, a novel method that utilizes efficient suffix trees to cache long token sequences from prompts and previous outputs. By adaptively speculating more tokens when acceptance likelihood is high and fewer when it is low, SuffixDecoding effectively exploits opportunities for longer speculations while conserving computation when those opportunities are limited. Evaluations on agentic benchmarks, including SWE-Bench and Text-to-SQL, demonstrate that SuffixDecoding achieves speedups of up to 5.3, outperforming state-of-the-art methods--2.8 faster than model-based approaches like EAGLE-2/3 and 1.9 faster than model-free approaches such as Token Recycling.

Reinforcement learning (RL) in partially observable settings is challenging because the agent's observations are not Markov. Recently proposed methods can learn variable-order Markov models of the underlying process but have steep memory requirements and are sensitive to aliasing between observation histories due to sensor noise. This paper proposes dynamic-depth context tree weighting (D2-CTW), a model-learning method that addresses these limitations. D2-CTW dynamically expands a suffix tree while ensuring that the size of the model, but not its depth, remains bounded. We show that D2-CTW approximately matches the performance of state-of-the-art alternatives at stochastic time-series prediction while using at least an order of magnitude less memory. We also apply D2-CTW to model-based RL, showing that, on tasks that require memory of past observations, D2-CTW can learn without prior knowledge of a good state representation, or even the length of history upon which such a representation should depend.

Reinforcement learning(RL) post-training has become essential for aligning large language models (LLMs), yet its efficiency is increasingly constrained by the rollout phase, where long trajectories are generated token by token. We identify a major bottleneck:the long-tail distribution of rollout lengths, where a small fraction of long generations dominates wall clock time and a complementary opportunity; the availability of historical rollouts that reveal stable prompt level patterns across training epochs. Motivated by these observations, we propose DAS, a Distribution Aware Speculative decoding framework that accelerates RL rollouts without altering model outputs. DAS integrates two key ideas: an adaptive, nonparametric drafter built from recent rollouts using an incrementally maintained suffix tree, and a length aware speculation policy that allocates more aggressive draft budgets to long trajectories that dominate makespan. This design exploits rollout history to sustain acceptance while balancing base and token level costs during decoding. Experiments on math and code reasoning tasks show that DAS reduces rollout time up to 50% while preserving identical training curves, demonstrating that distribution-aware speculative decoding can significantly accelerate RL post training without compromising learning quality.

We present SuffixDecoding, a novel model-free approach to accelerating large language model (LLM) inference through speculative decoding. Unlike existing methods that rely on draft models or specialized decoding heads, SuffixDecoding leverages suffix trees built from previously generated outputs to efficiently predict candidate token sequences. Our approach enables flexible tree-structured speculation without the overhead of maintaining and orchestrating additional models. SuffixDecoding builds and dynamically updates suffix trees to capture patterns in the generated text, using them to construct speculation trees through a principled scoring mechanism based on empirical token frequencies. SuffixDecoding requires only CPU memory which is plentiful and underutilized on typical LLM serving nodes. We demonstrate that SuffixDecoding achieves competitive speedups compared to model-based approaches across diverse workloads including open-domain chat, code generation, and text-to-SQL tasks. For open-ended chat and code generation tasks, SuffixDecoding achieves up to $1.4\times$ higher output throughput than SpecInfer and up to $1.1\times$ lower time-per-token (TPOT) latency. For a proprietary multi-LLM text-to-SQL application, SuffixDecoding achieves up to $2.9\times$ higher output throughput and $3\times$ lower latency than speculative decoding. Our evaluation shows that SuffixDecoding maintains high acceptance rates even with small reference corpora of 256 examples, while continuing to improve performance as more historical outputs are incorporated.

Reinforcement learning (RL) in partially observable settings is challenging because the agent's observations are not Markov. Recently proposed methods can learn variable-order Markov models of the underlying process but have steep memory requirements and are sensitive to aliasing between observation histories due to sensor noise. This paper proposes dynamic-depth context tree weighting (D2-CTW), a model-learning method that addresses these limitations. D2-CTW dynamically expands a suffix tree while ensuring that the size of the model, but not its depth, remains bounded. We show that D2-CTW approximately matches the performance of state-of-the-art alternatives at stochastic time-series prediction while using at least an order of magnitude less memory. We also apply D2-CTW to model-based RL, showing that, on tasks that require memory of past observations, D2-CTW can learn without prior knowledge of a good state representation, or even the length of history upon which such a representation should depend.

We present a neural sequence model designed specifically for symbolic music. The model is based on a learned edit distance mechanism which generalises a classic recursion from computer science, leading to a neural dynamic program. Repeated motifs are detected by learning the transformations between them. We represent the arising computational dependencies using a novel data structure, the edit tree; this perspective suggests natural approximations which afford the scaling up of our otherwise cubic time algorithm. We demonstrate our model on real and synthetic data; in all cases it outperforms a strong stacked long short-term memory benchmark.

Reinforcement learning (RL) in partially observable settings is challenging because the agent’s observations are not Markov. Recently proposed methods can learn variable-order Markov models of the underlying process but have steep memory requirements and are sensitive to aliasing between observation histories due to sensor noise. This paper proposes dynamic-depth context tree weighting (D2-CTW), a model-learning method that addresses these limitations. D2-CTW dynamically expands a suffix tree while ensuring that the size of the model, but not its depth, remains bounded. We show that D2-CTW approximately matches the performance of state-of-the-art alternatives at stochastic time-series prediction while using at least an order of magnitude less memory. We also apply D2-CTW to model-based RL, showing that, on tasks that require memory of past observations, D2-CTW can learn without prior knowledge of a good state representation, or even the length of history upon which such a representation should depend.

When William Legrand finally decrypted the string, it did not seem to make much more sense than it did before. But at least it did sound more like natural language, and eventually guided the main character of Edgar Allan Poe's "The Gold-Bug"36 to discover the treasure he had been after. Legrand solved a substitution cipher using symbol frequencies. He first looked for the most frequent symbol and changed it into the most frequent letter of English, then similarly inferred the most frequent word, then punctuation marks, and so on. Both before and after 1843, the natural impulse when faced with some mysterious message has been to count frequencies of individual tokens or subassemblies in search of a clue. Perhaps one of the most intense and fascinating subjects for this kind of scrutiny have been biosequences. As soon as some such sequences became available, statistical analysts tried to link characters or blocks of characters to relevant biological functions.

Document clustering as an unsupervised approach extensively used to navigate, filter, summarize and manage large collection of document repositories like the World Wide Web (WWW). Recently, focuses in this domain shifted from traditional vector based document similarity for clustering to suffix tree based document similarity, as it offers more semantic representation of the text present in the document. In this paper, we compare and contrast two recently introduced approaches to document clustering based on suffix tree data model. The first is an Efficient Phrase based document clustering, which extracts phrases from documents to form compact document representation and uses a similarity measure based on common suffix tree to cluster the documents. The second approach is a frequent word/word meaning sequence based document clustering, it similarly extracts the common word sequence from the document and uses the common sequence/ common word meaning sequence to perform the compact representation, and finally, it uses document clustering approach to cluster the compact documents. These algorithms are using agglomerative hierarchical document clustering to perform the actual clustering step, the difference in these approaches are mainly based on extraction of phrases, model representation as a compact document, and the similarity measures used for clustering. This paper investigates the computational aspect of the two algorithms, and the quality of results they produced.

Despite a dramatic growth of power consumption inhouseholds, less attention has been paid to monitoring,analyzing and predicting energy usage. In this paper,we propose a framework to mine raw energy data bytransforming time series energy data into a symbol se-quence, and then extend a suffix tree data structure asan efficient representation to analyze global structuralpatterns. Then, we use a clustering algorithm to detectenergy pattern outliers which are far from their clustercentroids. To validate our approach, we use real powerdata collected from a smart apartment testbed duringtwo months.