arXiv subjects

John Cooper

Publications and source records attributed to John Cooper.

At least 19 recordsLinked to original sources

arXiv2026

WARP: Weight-Space Analysis for Recovering Training Data Portfolios

Foundation models are routinely released to the public, yet the data recipes used to train them -- such as domain mixture weights that determine how different sources are sampled -- are rarely disclosed. This creates an access asymmetry: researchers study the resulting models but lack visibility into the training distribution that produces them. Prior works for inferring training data, such as membership inference, detect at the level of individual samples and thus cannot characterize the global composition of the training corpus. We introduce WARP, a framework that recovers a fine-tuned model's training mixtures directly from its released weights. WARP interpolates between the base and fine-tuned models using model merging, generating pseudo-checkpoints that approximate the missing training trajectory and expose a geometric footprint of the training data in the weight space. From these simulated footprints, WARP extracts geometric features and maps them to domain proportions using either a parameter-free softmax readout or an MLP projector trained on synthetic mixtures. In controlled experiments with BERT and GPT-2, WARP recovers domain mixtures with an average MAE as low as 0.046 and 0.104 respectively, outperforming membership inference and a variant with access to the true training trajectory.

cs.LG↗

arXiv2026

Expressivity-Efficiency Tradeoffs for Hybrid Sequence Models

Hybrid sequence models--combining Transformer and state-space model layers--seek to gain the expressive versatility of attention as well as the computational efficiency of state-space model layers. Despite burgeoning interest in hybrid models, we lack a basic understanding of the settings where--and underlying mechanisms through which--they offer benefits over their constituent models. In this paper, we study this question, focusing on a broad family of core synthetic tasks. For this family of tasks, we prove the existence of fundamental limitations for non-hybrid models. Specifically, any Transformer or state-space model that solves the underlying task requires either a large number of parameters or a large working memory. On the other hand, for two prototypical tasks within this family--namely selective copying and associative recall--we construct hybrid models of small size and working memory that provably solve these tasks, thus achieving the best of both worlds. Our experimental evaluation empirically validates our theoretical findings. Importantly, going beyond the settings in our theoretical analysis, we empirically show that learned--rather than constructed--hybrids outperform non-hybrid models with up to 6x as many parameters. We additionally demonstrate that hybrid models exhibit stronger length generalization and out-of-distribution robustness than non-hybrids.

cs.LG↗

arXiv2026

Weight Updates as Activation Shifts: A Principled Framework for Steering

Activation steering promises to be an extremely parameter-efficient form of adaptation, but its effectiveness depends on critical design choices -- such as intervention location and parameterization -- that currently rely on empirical heuristics rather than a principled foundation. We establish a first-order equivalence between activation-space interventions and weight-space updates, deriving the conditions under which activation steering can replicate fine-tuning behavior. This equivalence yields a principled framework for steering design and identifies the post-block output as a theoretically-backed and highly expressive intervention site. We further explain why certain intervention locations outperform others and show that weight updates and activation updates play distinct, complementary functional roles. This analysis motivates a new approach -- joint adaptation -- that trains in both spaces simultaneously. Our post-block steering achieves accuracy within 0.2%-0.9%$ of full-parameter tuning, on average across tasks and models, while training only 0.04% of model parameters. It consistently outperforms prior activation steering methods such as ReFT and PEFT approaches including LoRA, while using significantly fewer parameters. Finally, we show that joint adaptation often surpasses the performance ceilings of weight and activation updates in isolation, introducing a new paradigm for efficient model adaptation.

cs.LG↗

arXiv2026

Seeing Beyond Redundancy: Task Complexity's Role in Vision Token Specialization in VLLMs

Vision capabilities in vision large language models (VLLMs) have consistently lagged behind their linguistic capabilities. In particular, numerous benchmark studies have demonstrated that VLLMs struggle when fine-grained visual information or spatial reasoning is required. However, we do not yet understand exactly why VLLMs struggle so much with these tasks relative to others. Some works have focused on visual redundancy as an explanation, where high-level visual information is uniformly spread across numerous tokens and specific, fine-grained visual information is discarded. In this work, we investigate this premise in greater detail, seeking to better understand exactly how various types of visual information are processed by the model and what types of visual information are discarded. To do so, we introduce a simple synthetic benchmark dataset that is specifically constructed to probe various visual features, along with a set of metrics for measuring visual redundancy, allowing us to better understand the nuances of their relationship. Then, we explore fine-tuning VLLMs on a number of complex visual tasks to better understand how redundancy and compression change based upon the complexity of the data that a model is trained on. We find that there is a connection between task complexity and visual compression, implying that having a sufficient ratio of high complexity visual data is crucial for altering the way that VLLMs distribute their visual representation and consequently improving their performance on complex visual tasks. We hope that this work will provide valuable insights for training the next generation of VLLMs.

cs.CV↗

arXiv2025

Fully Collective Superradiant Lasing with Vanishing Sensitivity to Cavity Length Vibrations

To date, realization of a continuous-wave active atomic clock has been elusive primarily due to parasitic heating from spontaneous emission while repumping the atoms. Here, we propose a solution to this problem by replacing the random emission with coupling to an auxiliary cavity, making repumping a fully collective process. While it is known that collective two-level models do not possess a generic lasing threshold, we show this restriction is overcome with multi-level atoms since collective pumping and decay can be performed on distinct transitions. Using relevant atomic parameters, we find this system is capable of producing an $\mathcal{O}$(100 $\mu$Hz)-linewidth continuous-wave superradiant laser. Our principal result is the potential for an operating regime with cavity length vibration sensitivity below $\mathcal{O}(10^{-14} / g)$, including a locus of parameter values where it completely vanishes even at steady-state.

quant-ph↗

arXiv2025

R&B: Domain Regrouping and Data Mixture Balancing for Efficient Foundation Model Training

Data mixing strategies have successfully reduced the costs involved in training language models. While promising, such methods suffer from two flaws. First, they rely on predetermined data domains (e.g., data sources, task types), which may fail to capture critical semantic nuances, leaving performance on the table. Second, these methods scale with the number of domains in a computationally prohibitive way. We address these challenges via R&B, a framework that re-partitions training data based on semantic similarity (Regroup) to create finer-grained domains, and efficiently optimizes the data composition (Balance) by leveraging a Gram matrix induced by domain gradients obtained throughout training. Unlike prior works, it removes the need for additional compute to obtain evaluation information such as losses or gradients. We analyze this technique under standard regularity conditions and provide theoretical insights that justify R&B's effectiveness compared to non-adaptive mixing approaches. Empirically, we demonstrate the effectiveness of R&B on five diverse datasets ranging from natural language to reasoning and multimodal tasks. With as little as 0.01% additional compute overhead, R&B matches or exceeds the performance of state-of-the-art data mixing strategies.

cs.LG↗

arXiv2025

Foundation Models for Remote Sensing: An Analysis of MLLMs for Object Localization

Multimodal large language models (MLLMs) have altered the landscape of computer vision, obtaining impressive results across a wide range of tasks, especially in zero-shot settings. Unfortunately, their strong performance does not always transfer to out-of-distribution domains, such as earth observation (EO) imagery. Prior work has demonstrated that MLLMs excel at some EO tasks, such as image captioning and scene understanding, while failing at tasks that require more fine-grained spatial reasoning, such as object localization. However, MLLMs are advancing rapidly and insights quickly become out-dated. In this work, we analyze more recent MLLMs that have been explicitly trained to include fine-grained spatial reasoning capabilities, benchmarking them on EO object localization tasks. We demonstrate that these models are performant in certain settings, making them well suited for zero-shot scenarios. Additionally, we provide a detailed discussion focused on prompt selection, ground sample distance (GSD) optimization, and analyzing failure cases. We hope that this work will prove valuable as others evaluate whether an MLLM is well suited for a given EO localization task and how to optimize it.

cs.CV↗

arXiv2025

Evaluating Sample Utility for Efficient Data Selection by Mimicking Model Weights

Large-scale web-crawled datasets contain noise, bias, and irrelevant information, necessitating data selection techniques. Existing methods depend on hand-crafted heuristics, downstream datasets, or require expensive influence-based computations -- all of which limit scalability and introduce unwanted data dependencies. To address this, we introduce the Mimic Score, a simple and geometry-based data-quality metric that evaluates utility by measuring alignment between a sample's gradients and a target direction induced by a pre-trained reference model. This leverages readily available model weights, avoids needing validation datasets, and incurs minimal computational overheads. Building on this metric, we propose Grad-Mimic, a two-stage framework that re-weights samples online to accelerate training and aggregates sample utilities offline to construct effective data filters. Empirically, we show that using mimic scores to guide training improves data efficiency, accelerates convergence, yields consistent performance gains across six image datasets, and enhances CLIP models with 20.7% fewer training steps. Additionally, mimic score-based filters augment existing filtering techniques, enabling improved CLIP models trained with 4.7 million fewer samples.

cs.LG↗

arXiv2024

Weak-to-Strong Generalization Through the Data-Centric Lens

The weak-to-strong generalization phenomenon is the driver for important machine learning applications including highly data-efficient learning and, most recently, performing superalignment. While decades of research have resulted in numerous algorithms that produce strong empirical performance, understanding what aspects of data enable weak-to-strong generalization has been understudied. We propose a simple data-centric mechanism that characterizes weak-to-strong generalization: the overlap density. Intuitively, generalization tracks the number of points that contain overlaps, i.e., both easy patterns (learnable by a weak model) and challenging patterns (only learnable by a stronger model), as with such points, weak predictions can be used to learn challenging patterns by stronger models. We provide a practical overlap detection algorithm to find such points in datasets and leverage them to learn, among multiple sources of data, which to query when seeking to maximize overlap density and thereby enhance weak-to-strong generalization. We present a theoretical result showing that the generalization benefit is a function of the overlap density and a regret bound for our data selection algorithm. Empirically, we validate the mechanism and the overlap detection algorithm on a wide array of settings.

cs.LG↗

arXiv2024

Everything Everywhere All at Once: LLMs can In-Context Learn Multiple Tasks in Superposition

Large Language Models (LLMs) have demonstrated remarkable in-context learning (ICL) capabilities. In this study, we explore a surprising phenomenon related to ICL: LLMs can perform multiple, computationally distinct ICL tasks simultaneously, during a single inference call, a capability we term "task superposition". We provide empirical evidence of this phenomenon across various LLM families and scales and show that this phenomenon emerges even if we train the model to in-context learn one task at a time. We offer theoretical explanations that this capability is well within the expressive power of transformers. We also explore how LLMs internally compose task vectors during superposition. Furthermore, we show that larger models can solve more ICL tasks in parallel, and better calibrate their output distribution. Our findings offer insights into the latent capabilities of LLMs, further substantiate the perspective of "LLMs as superposition of simulators", and raise questions about the mechanisms enabling simultaneous task execution.

cs.LG↗

arXiv2024

MoRe Fine-Tuning with 10x Fewer Parameters

Parameter-efficient fine-tuning (PEFT) techniques have unlocked the potential to cheaply and easily specialize large pretrained models. However, the most prominent approaches, like low-rank adapters (LoRA), depend on heuristics or rules-of-thumb for their architectural choices -- potentially limiting their performance for new models and architectures. This limitation suggests that techniques from neural architecture search could be used to obtain optimal adapter architectures, but these are often expensive and difficult to implement. We address this challenge with Monarch Rectangular Fine-tuning (MoRe), a simple framework to search over adapter architectures that relies on the Monarch matrix class. Theoretically, we show that MoRe is more expressive than LoRA. Empirically, our approach is more parameter-efficient and performant than state-of-the-art PEFTs on a range of tasks and models, with as few as 5\% of LoRA's parameters.

cs.LG↗

arXiv2023

White paper on cybersecurity in the healthcare sector. The HEIR solution

The healthcare sector is increasingly vulnerable to cyberattacks due to its growing digitalization. Patient data, including medical records and financial information, are at risk, potentially leading to identity theft and patient safety concerns. The European Union and other organizations identify key areas for healthcare system improvement, yet the industry still grapples with inadequate security practices. In response, the HEIR project offers a comprehensive cybersecurity approach, promoting security features from various regulatory frameworks and introducing tools such as the Secure Healthcare Framework and Risk Assessment for Medical Applications (RAMA). These measures aim to enhance digital health security and protect sensitive patient data while facilitating secure data access and privacy-aware techniques. In a rapidly evolving threat landscape, HEIR presents a promising framework for healthcare cybersecurity.

cs.CR↗

arXiv2023

Speeding Up Squeezing with a Periodically Driven Dicke Model

We present a simple and effective method to create highly entangled spin states on a faster timescale than that of the commonly employed one-axis twisting (OAT) model. We demonstrate that by periodically driving the Dicke Hamiltonian at a resonance frequency, the system effectively becomes a two-axis countertwisting Hamiltonian which is known to quickly create Heisenberg limit scaled entangled states. For these states we show that simple quadrature measurements can saturate the ultimate precision limit for parameter estimation determined by the quantum Cram\'er-Rao bound. An example experimental realization of the periodically driven scheme is discussed with the potential to quickly generate momentum entanglement in a recently described experimental vertical cavity system. We analyze effects of collective dissipation in this vertical cavity system and find that our squeezing protocol can be more robust than the previous realization of OAT.

quant-ph↗

arXiv2022

Adiabatic Control of Decoherence-Free-Subspaces in an Open Collective System

We propose a method to adiabatically control an atomic ensemble using a decoherence-free subspace (DFS) within a dissipative cavity. We can engineer a specific eigenstate of the system's Lindblad jump operators by injecting a field into the cavity which deconstructively interferes with the emission amplitude of the ensemble. In contrast to previous adiabatic DFS proposals, our scheme creates a DFS in the presence of collective decoherence. We therefore have the ability to engineer states that have high multi-particle entanglements which may be exploited for quantum information science or metrology. We further demonstrate a more optimized driving scheme that utilizes the knowledge of possible diabatic evolution gained from the so-called adiabatic criteria. This allows us to evolve to a desired state with exceptionally high fidelity on a time scale that does not depend on the number of atoms in the ensemble. By engineering the DFS eigenstate adiabatically, our method allows for faster state preparation than previous schemes that rely on damping into a desired state solely using dissipation.

quant-ph↗

arXiv2021

Collective emission of an atomic beam into an off-resonant cavity mode

We study the collective emission of a beam of atomic dipoles into an optical cavity. Our focus lies on the effect of a finite detuning between the atomic transition frequency and the cavity resonance frequency. By developing a theoretical description of the coupled atom-cavity dynamics we analyze the stationary atomic configurations including a superradiant phase where the atoms undergo continuous monochromatic collective emission. In addition, we derive an analytical formula for the cavity pulling coefficient which characterizes the displacement of the emission frequency towards the cavity frequency. We find that the pulling is small if the cavity linewidth is much larger than the collective linewidth of the atomic beam. This regime is desired for building stable lasers because the emission frequency is robust against cavity length fluctuations. Furthermore, we investigate the stability of the atomic phases and compare our theoretical predictions with numerical results. Remarkably, we also find polychromatic emission regimes, where the spectrum has several frequency components while the light output is still superradiant.

quant-ph↗

arXiv2021

Superradiant emission of a thermal atomic beam into an optical cavity

We theoretically analyze the collective dynamics of a thermal beam of atomic dipoles that couple to a single mode when traversing an optical cavity. For this setup we derive a semiclassical model and determine the onset of superradiant emission and its stability. We derive analytical expressions for the linewidth of the emitted light and compare them with numerical simulations. In addition, we find and predict two different superradiant phases; a steady-state superradiant phase and a multi-component superradiant phase. In the latter case we observe sidebands in the frequency spectrum that can be calculated using a stability analysis of the amplitude mode of the collective dipole. We show that both superradiant phases are robust against free-space spontaneous emission and $T_2$ dephasing processes.

quant-ph↗

arXiv2020

Regular and bistable steady-state superradiant phases of an atomic beam traversing an optical cavity

We investigate the different photon emission regimes created by a preexcited and collimated atomic beam passing through a single mode of an optical cavity. In the regime where the cavity degrees of freedom can be adiabatically eliminated, we find that the atoms undergo superradiant emission when the collective linewidth exceeds the transit-time broadening. We analyze the case where the atomic beam direction is slanted with respect to the cavity axis. For this situation, we find that a phase of continuous light emission similar to steady-state superradiance is established providing the tilt of the atomic beam is sufficiently small. However, if the atoms travel more than half a wavelength along the cavity axis during one transit time we predict a dynamical phase transition to a new bistable superradiant regime. In this phase the atoms undergo collective spontaneous emission with a frequency that can be either blue or red detuned from the free-space atomic resonance. We analyze the different superradiant regimes and the quantum critical crossover boundaries. In particular we find the spectrum of the emitted light and show that the linewidth exhibits features of a critical scaling close to the phase boundaries.

quant-ph↗

arXiv2018

Dynamical phase transitions to optomechanical superradiance

We theoretically analyze superradiant emission of light from an ultracold gas of bosonic atoms confined in a bad cavity. A metastable dipolar transition of the atoms couples to the cavity field and is incoherently pumped, the mechanical effects of cavity-atom interactions tend to order the atoms in the periodic cavity potential. By means of a mean-field model we determine the conditions on the cavity parameters and pump rate that lead to the buildup of a stable macroscopic dipole emitting coherent light. We show that this occurs when the superradiant decay rate and the pump rate exceed threshold values of the order of the photon recoil energy. Above these thresholds superradiant emission is accompanied by the formation of stable matter-wave gratings that diffract the emitted photons. Outside of this regime, instead, the optomechanical coupling can give rise to dephasing or chaos, for which the emitted light is respectively incoherent or chaotic. These behaviors exhibit the features of a dynamical phase transitions and emerge from the interplay between global optomechanical interactions, quantum fluctuations, and noise.

quant-ph↗