arXiv subjects

Lawrence Liu

Publications and source records attributed to Lawrence Liu.

11 recordsLinked to original sources

arXiv2026

Making Alternative Data Work: Context-Augmented LLMs for Financial Forecasting

When forecasting a firm's future financial performance, alternative data - data collected from non-traditional sources such as consumer transactions, web traffic, and prediction markets - can provide timely signals about firms' operating activities and broader market conditions. These signals may reveal information that is not captured by traditional public sources and can therefore provide complementary information for forecasting firms' future financial performance. However, firm-level alternative data often have limited historical coverage, are relevant only to specific prediction targets or subsets of firms, and are distributed across numerous heterogeneous channels, making them difficult to incorporate flexibly into conventional forecasting approaches. Meanwhile, large language models (LLMs) can interpret instructions, learn from in-context examples, and generate predictions by combining heterogeneous information without task-specific parameter updates. Motivated by this potential flexibility, we investigate whether an LLM can forecast firm performance by integrating alternative data with other financial information through in-context learning. We propose a two-agent framework that first identifies the firms for which each alternative data channel is likely to be informative and then predicts revenue using firm- and channel-specific context. We evaluate the framework across four commercial alternative data channels. In our experiments, adding alternative data in context alongside other financial information improves the LLM's forecasting relative to either source alone, and these forecasts are more accurate than those of standard forecasting baselines. These findings suggest that LLMs provide a flexible and practical approach to integrating alternative data with heterogeneous financial information.

cs.AI↗

arXiv2026

LACONIC: Length-Aware Constrained Reinforcement Learning for LLM

Reinforcement learning (RL) has enhanced the capabilities of large language models (LLMs) through reward-driven training. Nevertheless, this process can introduce excessively long responses, inflating inference latency and computational overhead. Prior length-control approaches typically rely on fixed heuristic reward shaping, which can misalign with the task objective and require brittle tuning. In this work, we propose LACONIC, a reinforcement learning method that enforces a target token budget during training. Specifically, we update policy models using an augmented objective that combines the task reward with a length-based cost. To balance brevity and task performance, the cost scale is adaptively adjusted throughout training. This yields robust length control while preserving task reward. We provide a theoretical guarantee that support the method. Across mathematical reasoning models and datasets, LACONIC preserves or improves pass@1 while reducing output length by over 50%. It maintains out-of-domain performance on general knowledge and multilingual benchmarks with 44% fewer tokens. Moreover, LACONIC integrates into standard RL-tuning with no inference changes and minimal deployment overhead.

cs.LG↗

arXiv2026

Privacy-Preserving Cohort Analytics for Personalized Health Platforms: A Differentially Private Framework with Stochastic Risk Modeling

Personalized health analytics increasingly rely on population benchmarks to provide contextual insights such as ''How do I compare to others like me?'' However, cohort-based aggregation of health data introduces nontrivial privacy risks, particularly in interactive and longitudinal digital platforms. Existing privacy frameworks such as $k$-anonymity and differential privacy provide essential but largely static guarantees that do not fully capture the cumulative, distributional, and tail-dominated nature of re-identification risk in deployed systems. In this work, we present a privacy-preserving cohort analytics framework that combines deterministic cohort constraints, differential privacy mechanisms, and synthetic baseline generation to enable personalized population comparisons while maintaining strong privacy protections. We further introduce a stochastic risk modeling approach that treats re-identification risk as a random variable evolving over time, enabling distributional evaluation through Monte Carlo simulation. Adapting quantitative risk measures from financial mathematics, we define Privacy Loss at Risk (P-VaR) to characterize worst-case privacy outcomes under realistic cohort dynamics and adversary assumptions. We validate our framework through system-level analysis and simulation experiments, demonstrating how privacy-utility tradeoffs can be operationalized for digital health platforms. Our results suggest that stochastic risk modeling complements formal privacy guarantees by providing interpretable, decision-relevant metrics for platform designers, regulators, and clinical informatics stakeholders.

cs.CR↗

arXiv2026

FlexAct: Why Learn when you can Pick?

Learning activation functions has emerged as a promising direction in deep learning, allowing networks to adapt activation mechanisms to task-specific demands. In this work, we introduce a novel framework that employs the Gumbel-Softmax trick to enable discrete yet differentiable selection among a predefined set of activation functions during training. Our method dynamically learns the optimal activation function independently of the input, thereby enhancing both predictive accuracy and architectural flexibility. Experiments on synthetic datasets show that our model consistently selects the most suitable activation function, underscoring its effectiveness. These results connect theoretical advances with practical utility, paving the way for more adaptive and modular neural architectures in complex learning scenarios.

cs.LG↗

arXiv2025

Degree 2 vertices in minimal prime graph complements

Minimal prime graphs are connected graphs on at least two vertices whose complements satisfy the following conditions: triangle-freeness, 3-colorability, and edge-maximality with respect to the latter two properties. These graphs are prime graphs (or Gruenberg-Kegel graphs) of finite solvable groups with the maximum number of Frobenius actions among their Sylow subgroups, and as such minimal prime graph complements have been shown to be highly structured, including, for instance, the presence of induced 5-cycles. It is also known that the minimum degree of minimal prime graph complements is 2. In this note, we show that the existence of a degree 2 vertex in a minimal prime graph complement determines its whole structure: it is simply a 5-cycle with three vertices, exactly two of which are adjacent to each other, being duplicated finitely often. In particular, such graphs belong to a class of graphs known as reseminant.

math.CO↗

arXiv2025

ARMOR: High-Performance Semi-Structured Pruning via Adaptive Matrix Factorization

Large language models (LLMs) present significant deployment challenges due to their immense computational and memory requirements. While semi-structured pruning, particularly 2:4 sparsity, offers a path to practical hardware acceleration, existing methods often incur substantial performance degradation. To bridge this gap, we introduce ARMOR: (Adaptive Representation with Matrix-factORization), a novel one-shot post-training pruning algorithm. Instead of directly pruning weights, ARMOR factorizes each weight matrix into a 2:4 sparse core wrapped by two low-overhead, block diagonal matrices. These wrappers act as efficient pre and post-transformation error correctors, offering greater flexibility to preserve model quality compared to conventional 2:4 pruning techniques. The sparse core and block diagonal wrappers are chosen through a block coordinate descent algorithm that minimizes a layer-wise proxy loss. We theoretically prove this optimization is guaranteed to converge to a solution with a proxy loss less than or equal to state-of-the-art pruning algorithms. Experiments on Llama (Touvron et al., 2023; Dubey et al., 2024) and Qwen (Yang et al., 2025) model families demonstrate that ARMOR consistently and significantly outperforms state-of-the-art 2:4 pruning methods across a wide range of downstream tasks and perplexity evaluations. ARMOR achieves this superior performance while retaining the inference speedups and substantial memory usage reductions of 2:4 pruning, establishing a more effective trade-off between model compression and task accuracy

cs.LG↗

arXiv2025

NoWag: A Unified Framework for Shape Preserving Compression of Large Language Models

Large language models (LLMs) exhibit remarkable performance across various natural language processing tasks but suffer from immense computational and memory demands, limiting their deployment in resource-constrained environments. To address this challenge, we propose NoWag (Normalized Weight and Activation Guided Compression), a unified framework for one-shot shape preserving compression algorithms. We apply NoWag to compress Llama-2 (7B, 13B, 70B) and Llama-3 (8B, 70B) models using two popular shape-preserving techniques: vector quantization (NoWag-VQ) and unstructured/semi-structured pruning (NoWag-P). Our results show that NoWag-VQ significantly outperforms state-of-the-art one-shot vector quantization methods, while NoWag-P performs competitively against leading pruning techniques. These findings highlight underlying commonalities between these compression paradigms and suggest promising directions for future research. Our code is available at https://github.com/LawrenceRLiu/NoWag

cs.LG↗

arXiv2024

Emerging Jordan blocks in the two-dimensional Potts and loop models at generic $Q$

It was recently suggested -- based on general self-consistency arguments as well as results from the bootstrap (arXiv:2005.07708, arXiv:2007.11539, arXiv:2007.04190) -- that the CFT describing the $Q$-state Potts model is logarithmic for generic values of $Q$, with rank-two Jordan blocks for $L_0$ and ${\mkern 1.5mu\overline{\mkern-1.5mu L\mkern-1.5mu}\mkern 1.5mu}_0$ in many sectors of the theory. This is despite the well-known fact that the lattice transfer matrix (or Hamiltonian) is diagonalizable in (arbitrary) finite size. While the emergence of Jordan blocks only in the limit $L\to\infty$ is perfectly possible conceptually, diagonalizability in finite size makes the measurement of logarithmic couplings (whose values are analytically predicted in arXiv:2007.11539, arXiv:2007.04190) very challenging. This problem is solved in the present paper (which can be considered a companion to arXiv:2007.11539), and the conjectured logarithmic structure of the CFT confirmed in detail by the study of the lattice model and associated "emerging Jordan blocks."

math-ph↗

arXiv2024

Emerging Jordan forms, with applications to critical statistical models and conformal field theory

Two novel frameworks for handling mathematical and physical problems are introduced. The first, the emerging Jordan form, generalizes the concept of the Jordan canonical form, a well-established tool of linear algebra. The second, dual Jordan quantum physics, generalizes the framework of quantum physics to one in which the hermiticity postulate is considerably relaxed. These frameworks are then used to resolve some long-outstanding problems in theoretical physics, coming from critical statistical models and conformal field theory. I describe these problems and the difficulties involved in finding satisfactory solutions, then show how the concepts of emerging Jordan forms and dual Jordan quantum physics are naturally suited to overcoming these difficulties. Although their applications in this work are limited in scope to rather specific problems, the frameworks themselves are completely general, and I describe ways in which they may be used in other areas of mathematics and physics. Several appendices close the work, which include improvements to a widely used computational algorithm and corrections to some published data.

math-ph↗

arXiv2021

The action of the Virasoro algebra in the two-dimensional Potts and loop models at generic $Q$

The spectrum of conformal weights for the CFT describing the two-dimensional critical $Q$-state Potts model (or its close cousin, the dense loop model) has been known for more than 30 years. However, the exact nature of the corresponding $\hbox{Vir}\otimes\overline{\hbox{Vir}}$ representations has remained unknown up to now. Here, we solve the problem for generic values of $Q$. This is achieved by a mixture of different techniques: a careful study of "Koo--Saleur generators" [arXiv:hep-th/9312156], combined with measurements of four-point amplitudes, on the numerical side, and OPEs and the four-point amplitudes recently determined using the "interchiral conformal bootstrap" in [arXiv:2005.07258] on the analytical side. We find that null-descendants of diagonal fields having weights $(h_{r,1},h_{r,1})$ (with $r\in \mathbb{N}^*$) are truly zero, so these fields come with simple $\hbox{Vir}\otimes\overline{\hbox{Vir}}$ ("Kac") modules. Meanwhile, fields with weights $(h_{r,s},h_{r,-s})$ and $(h_{r,-s},h_{r,s})$ (with $r,s\in\mathbb{N}^*$) come in indecomposable but not fully reducible representations mixing four simple $\hbox{Vir}\otimes\overline{\hbox{Vir}}$ modules with a familiar "diamond" shape. The "top" and "bottom" fields in these diamonds have weights $(h_{r,-s},h_{r,-s})$, and form a two-dimensional Jordan cell for $L_0$ and $\bar{L}_0$. This establishes, among other things, that the Potts-model CFT is logarithmic for $Q$ generic. Unlike the case of non-generic (root of unity) values of $Q$, these indecomposable structures are not present in finite size, but we can nevertheless show from the numerical study of the lattice model how the rank-two Jordan cells build up in the infinite-size limit.

hep-th↗

arXiv2019

Ergodicity, eigenstate thermalization, and the foundations of statistical mechanics in quantum and classical systems

Boltzmann's ergodic hypothesis furnishes a possible explanation for the emergence of statistical mechanics in the framework of classical physics. In quantum mechanics, the Eigenstate Thermalization Hypothesis (ETH) is instead generally considered as a possible route to thermalization. This is because the notion of ergodicity itself is vague in the quantum world and it is often simply taken as a synonym for thermalization. Here we show, in an elementary way, that when quantum ergodicity is properly defined, it is, in fact, equivalent to ETH. In turn, ergodicity is equivalent to thermalization, thus implying the equivalence of thermalization and ETH. This result previously appeared in [De Palma et al., Phys. Rev. Lett. 115, 220401 (2015)], but becomes particularly clear in the present context. We also show that it is possible to define a classical analogue of ETH which is implicitly assumed to be satisfied when constructing classical statistical mechanics. Classical and quantum statistical mechanics are built according to the familiar standard prescription. This prescription, however, is ontologically justified only in the quantum world.

cond-mat.stat-mech↗