arXiv subjects

Naomi Sagan

Publications and source records attributed to Naomi Sagan.

10 recordsLinked to original sources

arXiv2026

M*: A Modular, Extensible, Serving System for Multimodal Models

We are entering a new era of composite model architectures that integrate diverse components such as vision encoders, language backbones, diffusion and flow heads, audio codecs, action generators, and world-model predictors. Such architectures underpin a broad class of multimodal models, including unified multimodal models, omni models, speech-language models, vision-language-action policies, and world models. However, existing model serving frameworks were built on narrow assumptions about model structure, making them ill-suited to accommodate this new architectural diversity. Here we present M*, a universal serving system for efficient serving of composite AI models. M* represents models as dataflow graphs, processing requests spanning diverse modalities and tasks as traversals over these graphs. The core insight is a modular abstraction that supports arbitrary composition of model components, flexible placement onto a physical cluster, and model-agnostic optimizations within a distributed runtime. We call this abstraction the Walk Graph and show how it can concisely capture composite models from a broad range of families. We instantiate M* on representative models and find that it achieves, on average, 20% lower end-to-end latency than vLLM-Omni for text-to-image workloads on BAGEL, while delivering up to 2.9x lower real-time factor and 2.7x higher throughput for text-to-speech workloads on Qwen3-Omni. M* also outperforms the V-JEPA 2-AC rollout baseline for robotic planning by up to 12.5x. Thus, our work paves the road towards more efficient serving of complex models with minimal developer effort.

cs.LG↗

arXiv2026

Learning When to Trust LLM Priors: A Validated Framework for Semantic Prior Integration

Large language models (LLMs) encode rich semantic knowledge that can be useful for supervised learning, but their outputs are unreliable as statistical priors: they may be noisy, misspecified, or hallucinated. Existing LLM-informed learning methods either trust such signals directly, leaving predictions vulnerable to unreliable LLM guidance, or restrict semantic integration to a single model class. We introduce Statsformer, a validated framework for learning when to trust LLM-derived semantic priors in supervised statistical learning. Statsformer maps LLM-derived feature scores into a family of learner-specific prior-injection mechanisms across a heterogeneous library of linear and nonlinear predictors. It then uses out-of-fold validation to adaptively calibrate the influence of each prior-informed learner, allowing useful semantic information to improve prediction while attenuating weak, misspecified, or adversarial priors. This yields a guardrailed statistical learning system with an oracle-style guarantee: up to statistical error, the final predictor performs no worse than the best convex combination of its in-library candidates, including prior-free learners. Across diverse prediction tasks, informative LLM priors improve performance, while unreliable priors are automatically downweighted. These results position Statsformer as a reliability-oriented approach to LLM-informed statistical learning: rather than trusting LLM knowledge directly, it validates semantic priors against data before allowing them to influence the final predictor.

stat.ML↗

arXiv2026

The LZ78 Source

We study a family of processes generated according to sequential probability assignments induced by the LZ78 universal compressor. We characterize entropic and distributional properties such as their entropy and relative entropy rates, finite-state compressibility and log loss of their realizations, and the empirical distributions that they induce. Though not quite stationary, these sources are "almost stationary and ergodic;" similar to stationary and ergodic processes, they satisfy a Shannon-McMillan-Breiman-type property: the normalized log probability of their realizations converges almost surely to their entropy rate. Further, they are locally "almost i.i.d." in the sense that the finite-dimensional empirical distributions of their realizations converge almost surely to a deterministic i.i.d. law. However, unlike stationary ergodic sources, the finite-state compressibility of their realizations is almost surely strictly larger than their entropy rate by a "Jensen gap". We present simulations demonstrating the theoretical results. These sources allow to gauge the performance of sequential probability models, both classical and deep learning-based, on non-Markovian non-stationary data. As such, we apply realizations of the LZ78 source to the study of in-context learning in transformer models.

cs.IT↗

arXiv2026

When Should Humans Step In? Optimal Human Dispatching in AI-Assisted Decisions

AI systems increasingly assist human decision making by producing preliminary assessments of complex inputs. However, such AI-generated assessments can often be noisy or systematically biased, raising a central question: how should costly human effort be allocated to correct AI outputs where it matters the most for the final decision? We propose a general decision-theoretic framework for human-AI collaboration in which AI assessments are treated as factor-level signals and human judgments as costly information that can be selectively acquired. We consider cases where the optimal selection problem reduces to maximizing a reward associated with each candidate subset of factors, and turn policy design into reward estimation. We develop estimation procedures under both nonparametric and linear models, covering contextual and non-contextual selection rules. In the linear setting, the optimal rule admits a closed-form expression with a clear interpretation in terms of factor importance and residual variance. We apply our framework to AI-assisted peer review. Our approach substantially outperforms LLM-only predictions and achieves performance comparable to full human review while using only 20-30% of the human information. Across different selection rules, we find that simpler rules derived under linear models can significantly reduce computational cost without harming final prediction performance. Our results highlight both the value of human intervention and the efficiency of principled dispatching.

stat.ML↗

arXiv2025

A Family of LZ78-based Universal Sequential Probability Assignments

We propose and study a family of universal sequential probability assignments on individual sequences, based on the incremental parsing procedure of the Lempel-Ziv (LZ78) compression algorithm. We show that the normalized log loss under any of these models converges to the normalized LZ78 codelength, uniformly over all individual sequences. To establish the universality of these models, we consolidate a set of results from the literature relating finite-state compressibility to optimal log-loss under Markovian and finite-state models. We also consider some theoretical and computational properties of these models when viewed as probabilistic sources. Finally, we present experimental results showcasing the potential benefit of using this family -- as models and as sources -- for compression, generation, and classification.

cs.IT↗

arXiv2025

LLM-Lasso: A Robust Framework for Domain-Informed Feature Selection and Regularization

We introduce LLM-Lasso, a novel framework that leverages large language models (LLMs) to guide feature selection in Lasso $\ell_1$ regression. Unlike traditional methods that rely solely on numerical data, LLM-Lasso incorporates domain-specific knowledge extracted from natural language, enhanced through a retrieval-augmented generation (RAG) pipeline, to seamlessly integrate data-driven modeling with contextual insights. Specifically, the LLM generates penalty factors for each feature, which are converted into weights for the Lasso penalty using a simple, tunable model. Features identified as more relevant by the LLM receive lower penalties, increasing their likelihood of being retained in the final model, while less relevant features are assigned higher penalties, reducing their influence. Importantly, LLM-Lasso has an internal validation step that determines how much to trust the contextual knowledge in our prediction pipeline. Hence it addresses key challenges in robustness, making it suitable for mitigating potential inaccuracies or hallucinations from the LLM. In various biomedical case studies, LLM-Lasso outperforms standard Lasso and existing feature selection baselines, all while ensuring the LLM operates without prior access to the datasets. To our knowledge, this is the first approach to effectively integrate conventional feature selection techniques directly with LLM-based domain-specific reasoning.

cs.LG↗

arXiv2025

Information-computation trade-offs in non-linear transforms

In this work, we explore the interplay between information and computation in non-linear transform-based compression for broad classes of modern information-processing tasks. We first investigate two emerging nonlinear data transformation frameworks for image compression: Implicit Neural Representations (INRs) and 2D Gaussian Splatting (GS). We analyze their representational properties, behavior under lossy compression, and convergence dynamics. Our results highlight key trade-offs between INR's compact, resolution-flexible neural field representations and GS's highly parallelizable, spatially interpretable fitting, providing insights for future hybrid and compression-aware frameworks. Next, we introduce the textual transform that enables efficient compression at ultra-low bitrate regimes and simultaneously enhances human perceptual satisfaction. When combined with the concept of denoising via lossy compression, the textual transform becomes a powerful tool for denoising tasks. Finally, we present a Lempel-Ziv (LZ78) "transform", a universal method that, when applied to any member of a broad compressor family, produces new compressors that retain the asymptotic universality guarantees of the LZ78 algorithm. Collectively, these three transforms illuminate the fundamental trade-offs between coding efficiency and computational cost. We discuss how these insights extend beyond compression to tasks such as classification, denoising, and generative AI, suggesting new pathways for using non-linear transformations to balance resource constraints and performance.

cs.IT↗

arXiv2025

LZMidi: Compression-Based Symbolic Music Generation

Recent advances in symbolic music generation primarily rely on deep learning models such as Transformers, GANs, and diffusion models. While these approaches achieve high-quality results, they require substantial computational resources, limiting their scalability. We introduce LZMidi, a lightweight symbolic music generation framework based on a Lempel-Ziv (LZ78)-induced sequential probability assignment (SPA). By leveraging the discrete and sequential structure of MIDI data, our approach enables efficient music generation on standard CPUs with minimal training and inference costs. Theoretically, we establish universal convergence guarantees for our approach, underscoring its reliability and robustness. Compared to state-of-the-art diffusion models, LZMidi achieves competitive Frechet Audio Distance (FAD), Wasserstein Distance (WD), and Kullback-Leibler (KL) scores, while significantly reducing computational overhead - up to 30x faster training and 300x faster generation. Our results position LZMidi as a significant advancement in compression-based learning, highlighting how universal compression techniques can efficiently model and generate structured sequential data, such as symbolic music, with practical scalability and theoretical rigor.

cs.SD↗

arXiv2025

Universal Discrete Filtering with Lookahead or Delay

We consider the universal discrete filtering problem, where an input sequence generated by an unknown source passes through a discrete memoryless channel, and the goal is to estimate its components based on the output sequence with limited lookahead or delay. We propose and establish the universality of a family of schemes for this setting. These schemes are induced by universal Sequential Probability Assignments (SPAs), and inherit their computational properties. We show that the schemes induced by LZ78 are practically implementable and well-suited for scenarios with limited computational resources and latency constraints. In passing, we use some of the intermediate results to obtain upper and lower bounds that appear to be new, in the purely Bayesian setting, on the optimal filtering performance in terms, respectively, of the mutual information between the noise-free and noisy sequence, and the entropy of the noise-free sequence causally conditioned on the noisy one.

eess.SP↗

arXiv2024

Compressing Large Language Models using Low Rank and Low Precision Decomposition

The prohibitive sizes of Large Language Models (LLMs) today make it difficult to deploy them on memory-constrained edge devices. This work introduces $\rm CALDERA$ -- a new post-training LLM compression algorithm that harnesses the inherent low-rank structure of a weight matrix $\mathbf{W}$ by approximating it via a low-rank, low-precision decomposition as $\mathbf{W} \approx \mathbf{Q} + \mathbf{L}\mathbf{R}$. Here, $\mathbf{L}$ and $\mathbf{R}$ are low rank factors, and the entries of $\mathbf{Q}$, $\mathbf{L}$ and $\mathbf{R}$ are quantized. The model is compressed by substituting each layer with its $\mathbf{Q} + \mathbf{L}\mathbf{R}$ decomposition, and the zero-shot performance of the compressed model is evaluated. Additionally, $\mathbf{L}$ and $\mathbf{R}$ are readily amenable to low-rank adaptation, consequently enhancing the zero-shot performance. $\rm CALDERA$ obtains this decomposition by formulating it as an optimization problem $\min_{\mathbf{Q},\mathbf{L},\mathbf{R}}\lVert(\mathbf{Q} + \mathbf{L}\mathbf{R} - \mathbf{W})\mathbf{X}^\top\rVert_{\rm F}^2$, where $\mathbf{X}$ is the calibration data, and $\mathbf{Q}, \mathbf{L}, \mathbf{R}$ are constrained to be representable using low-precision formats. Theoretical upper bounds on the approximation error of $\rm CALDERA$ are established using a rank-constrained regression framework, and the tradeoff between compression ratio and model performance is studied by analyzing the impact of target rank and quantization bit budget. Results illustrate that compressing LlaMa-$2$ $7$B/$13B$/$70$B and LlaMa-$3$ $8$B models using $\rm CALDERA$ outperforms existing post-training LLM compression techniques in the regime of less than $2.5$ bits per parameter. The implementation is available at: https://github.com/pilancilab/caldera.

cs.LG↗