Institution profile

Howard Hughes Medical Institute

Academic institutionnorthamerica · us
Official website
Research library27linked papers
Opportunities0open roles
Selected work

Representative Papers

Probability-turbulence divergence: A tunable allotaxonometric instrument for comparing heavy-tailed categorical distributions

Aug 30, 2020arXiv.org

Comparing frequency distributions across systems or over time under heavy-tailed regimes poses challenges due to sensitivity to zero-probability events and insufficient discrimination of subtle shifts. Method: We propose Probability Turbulence Divergence (PTD), a tunable, robust, and interpretable normalized divergence measure. PTD unifies classical metrics—including $L^p$ norms, the Sørensen–Dice coefficient, and the Hellinger distance—by integrating rank-based turbulence modeling and zero-probability embedding. It is theoretically linked to Rényi/Tsallis entropies and ecological Hill numbers. Contribution/Results: PTD exhibits zero-probability robustness and fine-grained frequency sensitivity, smoothly degenerating into multiple standard distances. We introduce the allotaxonograph—a novel multi-scale visualization framework—for granular analysis of frequency dynamics. Experiments on bibliometric, social media, and ecological datasets demonstrate PTD’s superior sensitivity to minor frequency perturbations. An open-source implementation enables cross-domain, interpretable comparisons.

3 citationsRead paper

Conditioned Direct Feedback Alignment via Activity and Error Geometry

Jul 20, 2026

This work addresses the training instability in Direct Feedback Alignment (DFA) caused by anisotropy in either presynaptic activities or local error signals. To mitigate this issue, the authors propose a normalized DFA approach that conditionally normalizes both activities and error signals. They introduce a symmetric, block-wise dual-factor normalization framework, decoupling and empirically validating the independent contributions of activity and error conditioning for the first time. The method is theoretically supported by linearized spectral analysis and implemented via inverse second-moment preconditioning and Kronecker-factor approximations. Controlled experiments on MNIST and Fashion-MNIST demonstrate that activity conditioning alone yields performance gains of up to 40 percentage points, while error conditioning improves accuracy by 1.77–7.53 percentage points; combining both strategies provides further modest improvements.

0 citationsRead paper

STSBench: A Large-Scale Dataset for Modeling Neuronal Activity in the Dorsal Stream of Primate Visual Cortex

Jul 17, 2026

This study addresses the limited understanding of the computational mechanisms underlying the primate dorsal visual stream, hindered by the absence of large-scale neural datasets. To overcome this, we introduce STSBench, a high-throughput electrophysiological dataset comprising activity from over 2,000 neurons in the superior temporal sulcus of macaques while they viewed thousands of natural videos. This resource represents a nearly 50-fold increase in scale over existing datasets and provides the first large-scale benchmark for the dorsal stream. STSBench enables robust training and evaluation of neural encoding models and demonstrates its utility by successfully reconstructing visual inputs from neural responses, thereby highlighting its pivotal value for both computational neuroscience and brain-inspired artificial intelligence.

0 citationsRead paper

Feature leakage and the identifiability of direct-dependency entropy models of neural activity

Jun 01, 2026

This study addresses a critical limitation of maximum entropy models in neural activity modeling: their tendency to misattribute higher-order statistical effects to first-order mechanisms due to biases in input distributions, thereby conflating predictive performance with genuine computational rules. To disentangle distribution-dependent prediction from mechanistic identifiability, the authors introduce a suite of diagnostic approaches—state reweighting, conditional log-odds contrast, and temporal leakage control—combined with constrained maximum entropy modeling, information projection, and coskewness analysis. Simulations demonstrate that purely higher-order responses can pass first-order tests under the original distribution yet are correctly identified after reweighting. Applied to CA1 hippocampal data, approximately half of the apparently first-order units exhibit distribution sensitivity upon reweighting, significantly exceeding expectations from an additive null model, thus revealing the limitations of interpreting neural coding solely through entropy maximization.

0 citationsRead paper
Recent publications

Latest Papers

Conditioned Direct Feedback Alignment via Activity and Error Geometry

Jul 20, 2026

This work addresses the training instability in Direct Feedback Alignment (DFA) caused by anisotropy in either presynaptic activities or local error signals. To mitigate this issue, the authors propose a normalized DFA approach that conditionally normalizes both activities and error signals. They introduce a symmetric, block-wise dual-factor normalization framework, decoupling and empirically validating the independent contributions of activity and error conditioning for the first time. The method is theoretically supported by linearized spectral analysis and implemented via inverse second-moment preconditioning and Kronecker-factor approximations. Controlled experiments on MNIST and Fashion-MNIST demonstrate that activity conditioning alone yields performance gains of up to 40 percentage points, while error conditioning improves accuracy by 1.77–7.53 percentage points; combining both strategies provides further modest improvements.

0 citationsRead paper

STSBench: A Large-Scale Dataset for Modeling Neuronal Activity in the Dorsal Stream of Primate Visual Cortex

Jul 17, 2026

This study addresses the limited understanding of the computational mechanisms underlying the primate dorsal visual stream, hindered by the absence of large-scale neural datasets. To overcome this, we introduce STSBench, a high-throughput electrophysiological dataset comprising activity from over 2,000 neurons in the superior temporal sulcus of macaques while they viewed thousands of natural videos. This resource represents a nearly 50-fold increase in scale over existing datasets and provides the first large-scale benchmark for the dorsal stream. STSBench enables robust training and evaluation of neural encoding models and demonstrates its utility by successfully reconstructing visual inputs from neural responses, thereby highlighting its pivotal value for both computational neuroscience and brain-inspired artificial intelligence.

0 citationsRead paper

Feature leakage and the identifiability of direct-dependency entropy models of neural activity

Jun 01, 2026

This study addresses a critical limitation of maximum entropy models in neural activity modeling: their tendency to misattribute higher-order statistical effects to first-order mechanisms due to biases in input distributions, thereby conflating predictive performance with genuine computational rules. To disentangle distribution-dependent prediction from mechanistic identifiability, the authors introduce a suite of diagnostic approaches—state reweighting, conditional log-odds contrast, and temporal leakage control—combined with constrained maximum entropy modeling, information projection, and coskewness analysis. Simulations demonstrate that purely higher-order responses can pass first-order tests under the original distribution yet are correctly identified after reweighting. Applied to CA1 hippocampal data, approximately half of the apparently first-order units exhibit distribution sensitivity upon reweighting, significantly exceeding expectations from an additive null model, thus revealing the limitations of interpreting neural coding solely through entropy maximization.

0 citationsRead paper

Task Relevance Is Not Local Replaceability: A Two-Axis View of Channel Information

May 07, 2026

This work addresses a critical limitation in existing channel pruning methods, which conflate two orthogonal dimensions—task relevance and local substitutability—thereby constraining performance. For the first time, this study explicitly disentangles these concepts: task relevance quantifies a channel’s contribution to the target objective, while local substitutability measures whether its function can be compensated by other channels within the same layer. Theoretical analysis and empirical evidence demonstrate that these two properties rapidly decouple during training, with local substitutability emerging as a more reliable criterion for pruning. Through comprehensive validation—including input attribution, channel overlap analysis, task information metrics, residual gradient examination, and ablation studies—this approach consistently outperforms conventional pruning strategies across multiple architectures and benchmarks, including CIFAR-100 and ImageNet.

0 citationsRead paper