arXiv subjects

Matt Jones

Publications and source records attributed to Matt Jones.

18 recordsLinked to original sources

arXiv2026

Context Sensitivity Improves Human-Machine Visual Alignment

Modern machine learning models typically represent inputs as fixed points in a high-dimensional embedding space. While this approach has been proven powerful for a wide range of downstream tasks, it fundamentally differs from the way humans process information. Because humans are constantly adapting to their environment, they represent objects and their relationships in a highly context-sensitive manner. To address this gap, we propose a method for context-sensitive similarity computation from neural network embeddings, applied to modeling a triplet odd-one-out task with an anchor image serving as simultaneous context. Modeling context enables us to achieve up to a 15% improvement in odd-one-out accuracy over a context-insensitive model. We find that this improvement is consistent across both original and "human-aligned" vision foundation models.

cs.CV↗

arXiv2025

Catch Your Breath: Adaptive Computation for Self-Paced Sequence Production

Within the landscape of inference-time scaling methods for foundation models, a width-based approach to scaling -- which involves the insertion of tokens in the input stream to delay model responses -- offers a unique advantage by increasing model expressivity while remaining highly parallelizable at both training and inference. The existing literature on training models to utilize tokens relies on the standard cross-entropy objective in which the model output is read out and evaluated only at the final step of a pause sequence. This approach provides no mechanism for the model to regulate its own processing or to signal readiness to respond, treating the additional compute steps as a static barrier rather than a resource to be used adaptively. We propose a supervised loss, Catch Your Breath (CYB), framed as a sequential-decision problem, that trains a model to dynamically and autonomously scale the number of compute steps used for each input token. The model indicates the need for additional compute steps by emitting a special output, delaying its response via a pause. The model can abstain multiple times to obtain longer delays. Our experiments demonstrate that CYB significantly outperforms standard cross-entropy when introduced either in pretraining or fine-tuning, reducing perplexity and enhancing downstream accuracy with no additional computational or memory cost.

cs.CL↗

arXiv2024

More Experts Than Galaxies: Conditionally-overlapping Experts With Biologically-Inspired Fixed Routing

The evolution of biological neural systems has led to both modularity and sparse coding, which enables energy efficiency and robustness across the diversity of tasks in the lifespan. In contrast, standard neural networks rely on dense, non-specialized architectures, where all model parameters are simultaneously updated to learn multiple tasks, leading to interference. Current sparse neural network approaches aim to alleviate this issue but are hindered by limitations such as 1) trainable gating functions that cause representation collapse, 2) disjoint experts that result in redundant computation and slow learning, and 3) reliance on explicit input or task IDs that limit flexibility and scalability. In this paper we propose Conditionally Overlapping Mixture of ExperTs (COMET), a general deep learning method that addresses these challenges by inducing a modular, sparse architecture with an exponential number of overlapping experts. COMET replaces the trainable gating function used in Sparse Mixture of Experts with a fixed, biologically inspired random projection applied to individual input representations. This design causes the degree of expert overlap to depend on input similarity, so that similar inputs tend to share more parameters. This results in faster learning per update step and improved out-of-sample generalization. We demonstrate the effectiveness of COMET on a range of tasks, including image classification, language modeling, and regression, using several popular deep learning architectures.

cs.LG↗

arXiv2024

On Principles and Representations for Extended Contextuality

There has been recent interest in whether the concept of quantum contextuality can be extended to systems with disturbance or signaling while retaining the essential properties of standard contextuality. Dzhafarov and Kujala (arXiv:2302.11995) offer a provocative argument that the answer is always affirmative: If standard contextuality satisfies some principle that extended contextuality does not, then that principle must be `non-substantive' in that it depends on a superficial choice of representation. Their argument is based on a ``consistification'' procedure that maps all systems to nondisturbing ones and that preserves contextuality. This paper raises several objections to their argument, including that it neglects how substantive principles change their expression under a change of representation, that it begs the question by presuming the principles are based on non-substantive transformations, and that the argument applies universally to arbitrary extensions of contextuality.

quant-ph↗

arXiv2024

Bayesian Online Natural Gradient (BONG)

We propose a novel approach to sequential Bayesian inference based on variational Bayes (VB). The key insight is that, in the online setting, we do not need to add the KL term to regularize to the prior (which comes from the posterior at the previous timestep); instead we can optimize just the expected log-likelihood, performing a single step of natural gradient descent starting at the prior predictive. We prove this method recovers exact Bayesian inference if the model is conjugate. We also show how to compute an efficient deterministic approximation to the VB objective, as well as our simplified objective, when the variational distribution is Gaussian or a sub-family, including the case of a diagonal plus low-rank precision matrix. We show empirically that our method outperforms other online VB methods in the non-conjugate setting, such as online learning for neural networks, especially when controlling for computational costs.

stat.ML↗

arXiv2024

Outlier-robust Kalman Filtering through Generalised Bayes

We derive a novel, provably robust, and closed-form Bayesian update rule for online filtering in state-space models in the presence of outliers and misspecified measurement models. Our method combines generalised Bayesian inference with filtering methods such as the extended and ensemble Kalman filter. We use the former to show robustness and the latter to ensure computational efficiency in the case of nonlinear models. Our method matches or outperforms other robust filtering methods (such as those based on variational Bayes) at a much lower computational cost. We show this empirically on a range of filtering problems with outlier measurements, such as object tracking, state estimation in high-dimensional chaotic systems, and online learning of neural networks.

stat.ML↗

arXiv2024

Reawakening knowledge: Anticipatory recovery from catastrophic interference via structured training

We explore the training dynamics of neural networks in a structured non-IID setting where documents are presented cyclically in a fixed, repeated sequence. Typically, networks suffer from catastrophic interference when training on a sequence of documents; however, we discover a curious and remarkable property of LLMs finetuned sequentially in this setting: they exhibit anticipatory behavior, recovering from the forgetting on documents before encountering them again. This behavior occurs even though the documents are never presented in context together. The behavior emerges and becomes more robust as the architecture scales up its number of parameters. Through comprehensive experiments and visualizations, we demonstrate a new mechanism by which over-parametrized neural networks can recover from catastrophic interference and uncover new insights into training over-parameterized networks in cyclically structured environments.

cs.LG↗

arXiv2024

Noise misleads rotation invariant algorithms on sparse targets

It is well known that the class of rotation invariant algorithms are suboptimal even for learning sparse linear problems when the number of examples is below the "dimension" of the problem. This class includes any gradient descent trained neural net with a fully-connected input layer (initialized with a rotationally symmetric distribution). The simplest sparse problem is learning a single feature out of $d$ features. In that case the classification error or regression loss grows with $1-k/n$ where $k$ is the number of examples seen. These lower bounds become vacuous when the number of examples $k$ reaches the dimension $d$. We show that when noise is added to this sparse linear problem, rotation invariant algorithms are still suboptimal after seeing $d$ or more examples. We prove this via a lower bound for the Bayes optimal algorithm on a rotationally symmetrized problem. We then prove much lower upper bounds on the same problem for simple non-rotation invariant algorithms. Finally we analyze the gradient flow trajectories of many standard optimization algorithms in some simple cases and show how they veer toward or away from the sparse targets. We believe that our trajectory categorization will be useful in designing algorithms that can exploit sparse targets and our method for proving lower bounds will be crucial for analyzing other families of algorithms that admit different classes of invariances.

stat.ML↗

arXiv2023

Low-rank extended Kalman filtering for online learning of neural networks from streaming data

We propose an efficient online approximate Bayesian inference algorithm for estimating the parameters of a nonlinear function from a potentially non-stationary data stream. The method is based on the extended Kalman filter (EKF), but uses a novel low-rank plus diagonal decomposition of the posterior precision matrix, which gives a cost per step which is linear in the number of model parameters. In contrast to methods based on stochastic variational inference, our method is fully deterministic, and does not require step-size tuning. We show experimentally that this results in much faster (more sample efficient) learning, which results in more rapid adaptation to changing distributions, and faster accumulation of reward when used as part of a contextual bandit algorithm.

stat.ML↗

arXiv2022

Impossibility Theorem for Extending Contextuality to Disturbing Systems

Recently there has been much interest and progress in extending the definition of contextuality to systems with disturbance. We prove that such an endeavor cannot simultaneously satisfy the following principles: (1) any deterministic system is noncontextual; (2) discarding information cannot turn a noncontextual system into a contextual one; (3) classical post-processing cannot create contextuality; (4) the joint realization of two statistically independent noncontextual systems is noncontextual. We also prove the same result without principle 4, under a stronger version of principle 1.

quant-ph↗

arXiv2022

Extended depth-of-field light-sheet microscopy improves imaging of large volumes at high numerical aperture

Light-sheet microscopes must compromise between field of view, optical sectioning, resolution, and detection efficiency. High-numerical-aperture (NA) detection objective lenses provide high resolution but their narrow depth of field fails to capture effectively the fluorescence signal generated by the illumination light sheets, in imaging large volumes. Here, we present ExD-SPIM (extended depth-of-field selective-plane illumination microscopy), an improved light-sheet microscopy strategy that solves this limitation by extending the depth of field (DOF) of high-NA detection objectives to match the thickness of the illumination light sheet. This extension of the DOF uses a phase mask to axially stretch the point-spread function of the objective lens while largely preserving lateral resolution. This matching of the detection DOF to the illumination-sheet thickness increases total fluorescence collection, reduces background, and improves the overall signal-to-noise ratio (SNR). We demonstrate, through numerical simulations and imaging of bead phantoms as well as living animals, that ExD-SPIM increases the SNR by more than three-fold and dramatically reduces the rate of fluorescence photobleaching, when compared to a low-NA system with an equivalent depth of field. Compared to conventional high-NA detection, ExD-SPIM improves the signal sensitivity and volumetric coverage of whole-brain activity imaging, increasing the number of detected neurons by over a third.

physics.optics↗

arXiv2020

Robust priors for regularized regression

Induction benefits from useful priors. Penalized regression approaches, like ridge regression, shrink weights toward zero but zero association is usually not a sensible prior. Inspired by simple and robust decision heuristics humans use, we constructed non-zero priors for penalized regression models that provide robust and interpretable solutions across several tasks. Our approach enables estimates from a constrained model to serve as a prior for a more general model, yielding a principled way to interpolate between models of differing complexity. We successfully applied this approach to a number of decision and classification problems, as well as analyzing simulated brain imaging data. Models with robust priors had excellent worst-case performance. Solutions followed from the form of the heuristic that was used to derive the prior. These new algorithms can serve applications in data analysis and machine learning, as well as help in understanding how people transition from novice to expert performance.

cs.LG↗

arXiv2020

Single-objective selective-volume illumination microscopy enables high-contrast light-field imaging

The performance of light-field microscopy is improved by selectively illuminating the relevant subvolume of the specimen with a second objective lens [1-3]. Here we advance this approach to a single-objective geometry, using an oblique one-photon illumination path or two-photon illumination to accomplish selective-volume excitation. The elimination of the second orthogonally oriented objective to selectively excite the volume of interest simplifies specimen mounting; yet, this single-objective approach still reduces out-of-volume background, resulting in improvements in image contrast, effective resolution, and volume reconstruction quality. We validate our new approach through imaging live developing zebrafish, demonstrating the technology's ability to capture imaging data from large volumes synchronously with high contrast, while remaining compatible with standard microscope sample mounting.

physics.optics↗

arXiv2019

Relating Causal and Probabilistic Approaches to Contextuality

A primary goal in recent research on contextuality has been to extend this concept to cases of inconsistent connectedness, where observables have different distributions in different contexts. This article proposes a solution within the framework of probabilistic causal models, which extend hidden-variables theories, and then demonstrates an equivalence to the contextuality-by-default (CbD) framework. CbD distinguishes contextuality from direct influences of context on observables, defining the latter purely in terms of probability distributions. Here we take a causal view of direct influences, defining direct influence within any causal model as the probability of all latent states of the system in which a change of context changes the outcome of a measurement. Model-based contextuality (M-contextuality) is then defined as the necessity of stronger direct influences to model a full system than when considered individually. For consistently connected systems, M-contextuality agrees with standard contextuality. For general systems, it is proved that M-contextuality is equivalent to the property that any model of a system must contain "hidden influences", meaning direct influences that go in opposite directions for different latent states, or equivalently signaling between observers that carries no information. This criterion can be taken as formalizing the "no-conspiracy" principle that has been proposed in connection with CbD. M-contextuality is then proved to be equivalent to CbD-contextuality, thus providing a new interpretation of CbD-contextuality as the non-existence of a model for a system without hidden direct influences.

quant-ph↗

arXiv2017

Reinforcement Learning with Analogical Similarity to Guide Schema Induction and Attention

Research in analogical reasoning suggests that higher-order cognitive functions such as abstract reasoning, far transfer, and creativity are founded on recognizing structural similarities among relational systems. Here we integrate theories of analogy with the computational framework of reinforcement learning (RL). We propose a psychology theory that is a computational synergy between analogy and RL, in which analogical comparison provides the RL learning algorithm with a measure of relational similarity, and RL provides feedback signals that can drive analogical learning. Simulation results support the power of this approach.

cs.AI↗

arXiv2015

On Contextuality in Behavioral Data

Dzhafarov, Zhang, and Kujala (Phil. Trans. Roy. Soc. A 374, 20150099) reviewed several behavioral data sets imitating the formal design of the quantum-mechanical contextuality experiments. The conclusion was that none of these data sets exhibited contextuality if understood in the generalized sense proposed in Dzhafarov, Kujala, and Larsson (Found. Phys. 7, 762-782, 2015), while the traditional definition of contextuality does not apply to these data because they violate the condition of consistent connectedness (also known as marginal selectivity, no-signaling condition, no-disturbance principle, etc.). In this paper we clarify the relationship between (in)consistent connectedness and (non)contextuality, as well as between the traditional and extended definitions of (non)contextuality, using as an example the Clauser-Horn-Shimony-Holt (CHSH) inequalities originally designed for detecting contextuality in entangled particles.

q-bio.NC↗

arXiv2004

Holographic generation of micro-trap arrays for single atoms

We have generated multiple micron-sized optical dipole traps for neutral atoms using holographic techniques with a programmable liquid crystal spatial light modulator. The setup allows the storing of a single atom per trap, and the addressing and manipulation of individual trapping sites.

quant-ph↗

arXiv1999

How a strongly irreducible Heegaard splitting intersects a handlebody

It has been shown elsewhere that a strongly irreducible Heegaard splitting surface Q of a compact orientable 3-manifold M can, under reasonable side conditions, intersect a ball or a solid torus in M in only a few possible ways. Here we extend those results to describe how Q can intersect a handlebody in M.

math.GT↗