Eli Shlizerman

Kun Su, Kaizhi Qian, Eli Shlizerman et al.

Modeling sounds emitted from physical object interactions is critical for immersive perceptual experiences in real and virtual worlds. Traditional methods of impact sound synthesis use physics simulation to obtain a set of physics parameters that could represent and synthesize the sound. However, they require fine details of both the object geometries and impact locations, which are rarely available in the real world and can not be applied to synthesize impact sounds from common videos. On the other hand, existing video-driven deep learning-based approaches could only capture the weak correspondence between visual content and impact sounds since they lack of physics knowledge. In this work, we propose a physics-driven diffusion model that can synthesize high-fidelity impact sound for a silent video clip. In addition to the video content, we propose to use additional physics priors to guide the impact sound synthesis procedure. The physics priors include both physics parameters that are directly estimated from noisy real-world impact sound examples without sophisticated setup and learned residual parameters that interpret the sound environment via neural networks. We further implement a novel diffusion model with specific training and inference strategies to combine physics priors and visual information for impact sound synthesis. Experimental results show that our model outperforms several existing systems in generating realistic impact sounds. More importantly, the physics-based representations are fully interpretable and transparent, thus enabling us to perform sound editing flexibly.

Trung Le, Eli Shlizerman

Modeling neural population dynamics underlying noisy single-trial spiking activities is essential for relating neural observation and behavior. A recent non-recurrent method - Neural Data Transformers (NDT) - has shown great success in capturing neural dynamics with low inference latency without an explicit dynamical model. However, NDT focuses on modeling the temporal evolution of the population activity while neglecting the rich covariation between individual neurons. In this paper we introduce SpatioTemporal Neural Data Transformer (STNDT), an NDT-based architecture that explicitly models responses of individual neurons in the population across time and space to uncover their underlying firing rates. In addition, we propose a contrastive learning loss that works in accordance with mask modeling objective to further improve the predictive performance. We show that our model achieves state-of-the-art performance on ensemble level in estimating neural activities across four neural datasets, demonstrating its capability to capture autonomous and non-autonomous dynamics spanning different cortical regions while being completely agnostic to the specific behaviors at hand. Furthermore, STNDT spatial attention mechanism reveals consistently important subsets of neurons that play a vital role in driving the response of the entire population, providing interpretability and key insights into how the population of neurons performs computation.

Kun Su, Xiulong Liu, Eli Shlizerman

Video encompasses both visual and auditory data, creating a perceptually rich experience where these two modalities complement each other. As such, videos are a valuable type of media for the investigation of the interplay between audio and visual elements. Previous studies of audio-visual modalities primarily focused on either audio-visual representation learning or generative modeling of a modality conditioned on the other, creating a disconnect between these two branches. A unified framework that learns representation and generates modalities has not been developed yet. In this work, we introduce a novel framework called Vision to Audio and Beyond (VAB) to bridge the gap between audio-visual representation learning and vision-to-audio generation. The key approach of VAB is that rather than working with raw video frames and audio data, VAB performs representation learning and generative modeling within latent spaces. In particular, VAB uses a pre-trained audio tokenizer and an image encoder to obtain audio tokens and visual features, respectively. It then performs the pre-training task of visual-conditioned masked audio token prediction. This training strategy enables the model to engage in contextual learning and simultaneous video-to-audio generation. After the pre-training phase, VAB employs the iterative-decoding approach to rapidly generate audio tokens conditioned on visual features. Since VAB is a unified model, its backbone can be fine-tuned for various audio-visual downstream tasks. Our experiments showcase the efficiency of VAB in producing high-quality audio from video, and its capability to acquire semantic audio-visual features, leading to competitive results in audio-visual retrieval and classification.

Lu Mi, Trung Le, Tianxing He et al.

Neurons can display highly variable dynamics. While such variability presumably supports the wide range of behaviors generated by the organism, their gene expressions are relatively stable in the adult brain. This suggests that neuronal activity is a combination of its time-invariant identity and the inputs the neuron receives from the rest of the circuit. Here, we propose a self-supervised learning based method to assign time-invariant representations to individual neurons based on permutation-, and population size-invariant summary of population recordings. We fit dynamical models to neuronal activity to learn a representation by considering the activity of both the individual and the neighboring population. Our self-supervised approach and use of implicit representations enable robust inference against imperfections such as partial overlap of neurons across sessions, trial-to-trial variability, and limited availability of molecular (transcriptomic) labels for downstream supervised tasks. We demonstrate our method on a public multimodal dataset of mouse cortical neuronal activity and transcriptomic labels. We report > 35% improvement in predicting the transcriptomic subclass identity and > 20% improvement in predicting class identity with respect to the state-of-the-art.

Jinlin Xiang, Eli Shlizerman

When learning new tasks in a sequential manner, deep neural networks tend to forget tasks that they previously learned, a phenomenon called catastrophic forgetting. Class incremental learning methods aim to address this problem by keeping a memory of a few exemplars from previously learned tasks, and distilling knowledge from them. However, existing methods struggle to balance the performance across classes since they typically overfit the model to the latest task. In our work, we propose to address these challenges with the introduction of a novel methodology of Tangent Kernel for Incremental Learning (TKIL) that achieves class-balanced performance. The approach preserves the representations across classes and balances the accuracy for each class, and as such achieves better overall accuracy and variance. TKIL approach is based on Neural Tangent Kernel (NTK), which describes the convergence behavior of neural networks as a kernel function in the limit of infinite width. In TKIL, the gradients between feature layers are treated as the distance between the representations of these layers and can be defined as Gradients Tangent Kernel loss (GTK loss) such that it is minimized along with averaging weights. This allows TKIL to automatically identify the task and to quickly adapt to it during inference. Experiments on CIFAR-100 and ImageNet datasets with various incremental learning settings show that these strategies allow TKIL to outperform existing state-of-the-art methods.

Caleb Zheng, Jyotika Singh, Fang Tu et al.

In-Context Learning and Chain-of-Thought prompting improve reasoning in large language models (LLMs). These typically come at the cost of longer, more expensive prompts that may contain redundant information. Prompt compression based on pruning offers a practical solution, yet existing methods rely on sequential token removal which is computationally intensive. We present DiffuMask, a diffusion-based framework integrating hierarchical shot-level and token-level pruning signals, that enables rapid and parallel prompt pruning via iterative mask prediction. DiffuMask substantially accelerates the compression process via masking multiple tokens in each denoising step. It offers tunable control over retained content, preserving essential reasoning context and achieving up to 80\% prompt length reduction. Meanwhile, it maintains or improves accuracy across in-domain, out-of-domain, and cross-model settings. Our results show that DiffuMask provides a generalizable and controllable framework for prompt compression, facilitating faster and more reliable in-context reasoning in LLMs.

Ryan Vogt, Yang Zheng, Eli Shlizerman

Recurrent Neural Networks (RNN) are ubiquitous computing systems for sequences and multivariate time series data. While several robust architectures of RNN are known, it is unclear how to relate RNN initialization, architecture, and other hyperparameters with accuracy for a given task. In this work, we propose to treat RNN as dynamical systems and to correlate hyperparameters with accuracy through Lyapunov spectral analysis, a methodology specifically designed for nonlinear dynamical systems. To address the fact that RNN features go beyond the existing Lyapunov spectral analysis, we propose to infer relevant features from the Lyapunov spectrum with an Autoencoder and an embedding of its latent representation (AeLLE). Our studies of various RNN architectures show that AeLLE successfully correlates RNN Lyapunov spectrum with accuracy. Furthermore, the latent representation learned by AeLLE is generalizable to novel inputs from the same task and is formed early in the process of RNN training. The latter property allows for the prediction of the accuracy to which RNN would converge when training is complete. We conclude that representation of RNN through Lyapunov spectrum along with AeLLE provides a novel method for organization and interpretation of variants of RNN architectures.

Caleb Zheng, Eli Shlizerman

Diffusion models achieve remarkable performance across diverse generative tasks in computer vision, but their high computational cost remains a major barrier to deployment. Model pruning offers a promising way to reduce inference cost and enable lightweight models. However, pruning leads to quality drop due to reduced capacity. A key limitation of existing pruning approaches is that pruned models are finetuned using the same objective as the dense model (denoising score matching). Since the dense model is accessible during finetuning, it warrants a more effective approach for knowledge transfer from the dense to the pruned model. Motivated by this, we propose \textbf{2ndMatch} (\textbf{2ndM}), a general-purpose finetuning framework that introduces a \textbf{2nd}-order Jacobian ($J^{\top} J$) \textbf{M}atching loss inspired by Finite-Time Lyapunov Exponents. \textbf{2ndM} teaches the pruned model to mimic the sensitivity of the dense teacher, i.e., how to respond to small perturbations over time, through scalable random projections. The framework is architecture-agnostic and applies to both U-Net- and Transformer-based diffusion models. Experiments on CIFAR-10, CelebA, LSUN, ImageNet, and MSCOCO demonstrate that \textbf{2ndM} reduces the performance gap between pruned and dense models, substantially improving output quality.

Mingfei Chen, Yifan Wang, Zhengqin Li et al.

Prior works on 3D hand trajectory prediction are constrained by datasets that decouple motion from semantic supervision and by models that weakly link reasoning and action. To address these, we first present the EgoMAN dataset, a large-scale egocentric dataset for interaction stage-aware 3D hand trajectory prediction with 219K 6DoF trajectories and 3M structured QA pairs for semantic, spatial, and motion reasoning. We then introduce the EgoMAN model, a reasoning-to-motion framework that links vision-language reasoning and motion generation via a trajectory-token interface. Trained progressively to align reasoning with motion dynamics, our approach yields accurate and stage-aware trajectories with generalization across real-world scenes.

James Hazelden, Laura Driscoll, Eli Shlizerman et al.

In training a neural network with gradient descent (GD), each iteration induces a linear operator that governs first-order updates to a model's internal state variables. We define this operator as the Global Empirical Neural Tangent Kernel (NTK). In finite-width networks, the NTK is typically intractable to form, leading prior work to focus on restrictive settings such as tracking outputs only or taking infinite-width limits. Here, we study the structure of the NTK for a range of models. Formulating the model state as the solution to a single global implicit constraint, we derive the NTK as a product of two operators: K, accounting for immediate parameter-to-state interactions, and P, describing internal state-to-state dependencies. For a broad class of weight-based models, including RNNs and transformers, we prove a universal Kronecker-core theorem showing that K admits an exact, computable form given by the Gram matrix of weight-site variables. This core structure reveals that the NTK is structurally bottlenecked, constraining its effective rank and giving rise to a self-referential bias whereby GD preferentially learns within dominant modes of joint hidden and input activity. For recurrent models, we examine the spectrum of the NTK and show when it is biased and low-rank in space or time under the proposed decomposition. We further demonstrate that model dynamics at initialization bias the NTK, restricting learning and preventing task components from being learned effectively. Finally, we show that the NTK associated with a self-attention transformer is likewise structurally constrained to be low-rank. Overall, we show that the NTK possesses tractable structure that explains GD bias toward task solutions and the emergence of low-rank representations. To enable use of the NTK as a practical metric, we build kpflow, a library relying on randomized matrix-free numerical linear algebra.

Claudius Krause, Michele Faucci Giannelli, Gregor Kasieczka et al.

We present the results of the "Fast Calorimeter Simulation Challenge 2022" - the CaloChallenge. We study state-of-the-art generative models on four calorimeter shower datasets of increasing dimensionality, ranging from a few hundred voxels to a few tens of thousand voxels. The 31 individual submissions span a wide range of current popular generative architectures, including Variational AutoEncoders (VAEs), Generative Adversarial Networks (GANs), Normalizing Flows, Diffusion models, and models based on Conditional Flow Matching. We compare all submissions in terms of quality of generated calorimeter showers, as well as shower generation time and model size. To assess the quality we use a broad range of different metrics including differences in 1-dimensional histograms of observables, KPD/FPD scores, AUCs of binary classifiers, and the log-posterior of a multiclass classifier. The results of the CaloChallenge provide the most complete and comprehensive survey of cutting-edge approaches to calorimeter fast simulation to date. In addition, our work provides a uniquely detailed perspective on the important problem of how to evaluate generative models. As such, the results presented here should be applicable for other domains that use generative AI and require fast and faithful generation of samples in a large phase space.

Xiulong Liu, Kun Su, Eli Shlizerman

The content of visual and audio scenes is multi-faceted such that a video can be paired with various audio and vice-versa. Thereby, in video-to-audio generation task, it is imperative to introduce steering approaches for controlling the generated audio. While Video-to-Audio generation is a well-established generative task, existing methods lack such controllability. In this work, we propose VATT, a multi-modal generative framework that takes a video and an optional text prompt as input, and generates audio and optional textual description of the audio. Such a framework has two advantages: i) Video-to-Audio generation process can be refined and controlled via text which complements the context of visual information, and ii) The model can suggest what audio to generate for the video by generating audio captions. VATT consists of two key modules: VATT Converter, a LLM that is fine-tuned for instructions and includes a projection layer that maps video features to the LLM vector space; and VATT Audio, a transformer that generates audio tokens from visual frames and from optional text prompt using iterative parallel decoding. The audio tokens are converted to a waveform by pretrained neural codec. Experiments show that when VATT is compared to existing video-to-audio generation methods in objective metrics, it achieves competitive performance when the audio caption is not provided. When the audio caption is provided as a prompt, VATT achieves even more refined performance (lowest KLD score of 1.41). Furthermore, subjective studies show that VATT Audio has been chosen as preferred generated audio than audio generated by existing methods. VATT enables controllable video-to-audio generation through text as well as suggesting text prompts for videos through audio captions, unlocking novel applications such as text-guided video-to-audio generation and video-to-audio captioning.

Anna Wirth-Singh, Jinlin Xiang, Minho Choi et al.

Optical and hybrid convolutional neural networks (CNNs) recently have become of increasing interest to achieve low-latency, low-power image classification and computer vision tasks. However, implementing optical nonlinearity is challenging, and omitting the nonlinear layers in a standard CNN comes at a significant reduction in accuracy. In this work, we use knowledge distillation to compress modified AlexNet to a single linear convolutional layer and an electronic backend (two fully connected layers). We obtain comparable performance to a purely electronic CNN with five convolutional layers and three fully connected layers. We implement the convolution optically via engineering the point spread function of an inverse-designed meta-optic. Using this hybrid approach, we estimate a reduction in multiply-accumulate operations from 17M in a conventional electronic modified AlexNet to only 86K in the hybrid compressed network enabled by the optical frontend. This constitutes over two orders of magnitude reduction in latency and power consumption. Furthermore, we experimentally demonstrate that the classification accuracy of the system exceeds 93% on the MNIST dataset.

Qibin Liu, Chase Shimmin, Xiulong Liu et al.

We introduce a novel machine learning method developed for the fast simulation of calorimeter detector response, adapting vector-quantized variational autoencoder (VQ-VAE). Our model adopts a two-stage generation strategy: initially compressing geometry-aware calorimeter data into a discrete latent space, followed by the application of a sequence model to learn and generate the latent tokens. Extensive experimentation on the Calo-challenge dataset underscores the efficiency of our approach, showcasing a remarkable improvement in the generation speed compared with conventional method by a factor of 2000. Remarkably, our model achieves the generation of calorimeter showers within milliseconds. Furthermore, comprehensive quantitative evaluations across various metrics are performed to validate physics performance of generation.

Xiulong Liu, Anurag Kumar, Paul Calamia et al.

In mixed reality applications, a realistic acoustic experience in spatial environments is as crucial as the visual experience for achieving true immersion. Despite recent advances in neural approaches for Room Impulse Response (RIR) estimation, most existing methods are limited to the single environment on which they are trained, lacking the ability to generalize to new rooms with different geometries and surface materials. We aim to develop a unified model capable of reconstructing the spatial acoustic experience of any environment with minimum additional measurements. To this end, we present xRIR, a framework for cross-room RIR prediction. The core of our generalizable approach lies in combining a geometric feature extractor, which captures spatial context from panorama depth images, with a RIR encoder that extracts detailed acoustic features from only a few reference RIR samples. To evaluate our method, we introduce ACOUSTICROOMS, a new dataset featuring high-fidelity simulation of over 300,000 RIRs from 260 rooms. Experiments show that our method strongly outperforms a series of baselines. Furthermore, we successfully perform sim-to-real transfer by evaluating our model on four real-world environments, demonstrating the generalizability of our approach and the realism of our dataset.

Minho Choi, Jinlin Xiang, Anna Wirth-Singh et al.

Artificial neural networks (ANNs) have fundamentally transformed the field of computer vision, providing unprecedented performance. However, these ANNs for image processing demand substantial computational resources, often hindering real-time operation. In this paper, we demonstrate an optical encoder that can perform convolution simultaneously in three color channels during the image capture, effectively implementing several initial convolutional layers of a ANN. Such an optical encoding results in ~24,000 times reduction in computational operations, with a state-of-the art classification accuracy (~73.2%) in free-space optical system. In addition, our analog optical encoder, trained for CIFAR-10 data, can be transferred to the ImageNet subset, High-10, without any modifications, and still exhibits moderate accuracy. Our results evidence the potential of hybrid optical/digital computer vision system in which the optical frontend can pre-process an ambient scene to reduce the energy and latency of the whole computer vision system.

Elizabeth G. Campolongo, Yuan-Tang Chou, Ekaterina Govorkova et al.

Scientific discoveries are often made by finding a pattern or object that was not predicted by the known rules of science. Oftentimes, these anomalous events or objects that do not conform to the norms are an indication that the rules of science governing the data are incomplete, and something new needs to be present to explain these unexpected outliers. The challenge of finding anomalies can be confounding since it requires codifying a complete knowledge of the known scientific behaviors and then projecting these known behaviors on the data to look for deviations. When utilizing machine learning, this presents a particular challenge since we require that the model not only understands scientific data perfectly but also recognizes when the data is inconsistent and out of the scope of its trained behavior. In this paper, we present three datasets aimed at developing machine learning-based anomaly detection for disparate scientific domains covering astrophysics, genomics, and polar science. We present the different datasets along with a scheme to make machine learning challenges around the three datasets findable, accessible, interoperable, and reusable (FAIR). Furthermore, we present an approach that generalizes to future machine learning challenges, enabling the possibility of large, more compute-intensive challenges that can ultimately lead to scientific discovery.

Francis R. Willett, Jingyuan Li, Trung Le et al.

Speech brain-computer interfaces aim to decipher what a person is trying to say from neural activity alone, restoring communication to people with paralysis who have lost the ability to speak intelligibly. The Brain-to-Text Benchmark '24 and associated competition was created to foster the advancement of decoding algorithms that convert neural activity to text. Here, we summarize the lessons learned from the competition ending on June 1, 2024 (the top 4 entrants also presented their experiences in a recorded webinar). The largest improvements in accuracy were achieved using an ensembling approach, where the output of multiple independent decoders was merged using a fine-tuned large language model (an approach used by all 3 top entrants). Performance gains were also found by improving how the baseline recurrent neural network (RNN) model was trained, including by optimizing learning rate scheduling and by using a diphone training objective. Improving upon the model architecture itself proved more difficult, however, with attempts to use deep state space models or transformers not yet appearing to offer a benefit over the RNN baseline. The benchmark will remain open indefinitely to support further work towards increasing the accuracy of brain-to-text algorithms.

Mingfei Chen, Zijun Cui, Xiulong Liu et al.

3D spatial reasoning in dynamic, audio-visual environments is a cornerstone of human cognition yet remains largely unexplored by existing Audio-Visual Large Language Models (AV-LLMs) and benchmarks, which predominantly focus on static or 2D scenes. We introduce SAVVY-Bench, the first benchmark for 3D spatial reasoning in dynamic scenes with synchronized spatial audio. SAVVY-Bench is comprised of thousands of relationships involving static and moving objects, and requires fine-grained temporal grounding, consistent 3D localization, and multi-modal annotation. To tackle this challenge, we propose SAVVY, a novel training-free reasoning pipeline that consists of two stages: (i) Egocentric Spatial Tracks Estimation, which leverages AV-LLMs as well as other audio-visual methods to track the trajectories of key objects related to the query using both visual and spatial audio cues, and (ii) Dynamic Global Map Construction, which aggregates multi-modal queried object trajectories and converts them into a unified global dynamic map. Using the constructed map, a final QA answer is obtained through a coordinate transformation that aligns the global map with the queried viewpoint. Empirical evaluation demonstrates that SAVVY substantially enhances performance of state-of-the-art AV-LLMs, setting a new standard and stage for approaching dynamic 3D spatial reasoning in AV-LLMs.

Mingfei Chen, Israel D. Gebru, Ishwarya Ananthabhotla et al.

We introduce SoundVista, a method to generate the ambient sound of an arbitrary scene at novel viewpoints. Given a pre-acquired recording of the scene from sparsely distributed microphones, SoundVista can synthesize the sound of that scene from an unseen target viewpoint. The method learns the underlying acoustic transfer function that relates the signals acquired at the distributed microphones to the signal at the target viewpoint, using a limited number of known recordings. Unlike existing works, our method does not require constraints or prior knowledge of sound source details. Moreover, our method efficiently adapts to diverse room layouts, reference microphone configurations and unseen environments. To enable this, we introduce a visual-acoustic binding module that learns visual embeddings linked with local acoustic properties from panoramic RGB and depth data. We first leverage these embeddings to optimize the placement of reference microphones in any given scene. During synthesis, we leverage multiple embeddings extracted from reference locations to get adaptive weights for their contribution, conditioned on target viewpoint. We benchmark the task on both publicly available data and real-world settings. We demonstrate significant improvements over existing methods.

James Hazelden, Laura Driscoll, Eli Shlizerman et al.

Gradient Descent (GD) and its variants are the primary tool for enabling efficient training of recurrent dynamical systems such as Recurrent Neural Networks (RNNs), Neural ODEs and Gated Recurrent units (GRUs). The dynamics that are formed in these models exhibit features such as neural collapse and emergence of latent representations that may support the remarkable generalization properties of networks. In neuroscience, qualitative features of these representations are used to compare learning in biological and artificial systems. Despite recent progress, there remains a need for theoretical tools to rigorously understand the mechanisms shaping learned representations, especially in finite, non-linear models. Here, we show that the gradient flow, which describes how the model's dynamics evolve over GD, can be decomposed into a product that involves two operators: a Parameter Operator, K, and a Linearized Flow Propagator, P. K mirrors the Neural Tangent Kernel in feed-forward neural networks, while P appears in Lyapunov stability and optimal control theory. We demonstrate two applications of our decomposition. First, we show how their interplay gives rise to low-dimensional latent dynamics under GD, and, specifically, how the collapse is a result of the network structure, over and above the nature of the underlying task. Second, for multi-task training, we show that the operators can be used to measure how objectives relevant to individual sub-tasks align. We experimentally and theoretically validate these findings, providing an efficient Pytorch package, \emph{KPFlow}, implementing robust analysis tools for general recurrent architectures. Taken together, our work moves towards building a next stage of understanding of GD learning in non-linear recurrent models.

Hao Fang, Ryan A. Canfield, Tomohiro Ouchi et al.

Brain decoding aims to interpret and translate neural activity into behaviors. As such, it is imperative that decoding models are able to generalize across variations, such as recordings from different brain sites, distinct sessions, different types of behavior, and a variety of subjects. Current models can only partially address these challenges and warrant the development of pretrained neural transformer models capable to adapt and generalize. In this work, we propose RPNT - Robust Pretrained Neural Transformer, designed to achieve robust generalization through pretraining, which in turn enables effective finetuning given a downstream task. In particular, RPNT unique components include 1) Multidimensional rotary positional embedding (MRoPE) to aggregate experimental metadata such as site coordinates, session name and behavior types; 2) Context-based attention mechanism via convolution kernels operating on global attention to learn local temporal structures for handling non-stationarity of neural population activity; 3) Robust self-supervised learning (SSL) objective with uniform causal masking strategies and contrastive representations. We pretrained two separate versions of RPNT on distinct datasets a) Multi-session, multi-task, and multi-subject microelectrode benchmark; b) Multi-site recordings using high-density Neuropixel 1.0 probes. The datasets include recordings from the dorsal premotor cortex (PMd) and from the primary motor cortex (M1) regions of nonhuman primates (NHPs) as they performed reaching tasks. After pretraining, we evaluated the generalization of RPNT in cross-session, cross-type, cross-subject, and cross-site downstream behavior decoding tasks. Our results show that RPNT consistently achieves and surpasses the decoding performance of existing decoding models in all tasks.

Jinlin Xiang, Minho Choi, Yubo Zhang et al.

Hybrid Optical Neural Networks (ONNs, typically consisting of an optical frontend and a digital backend) offer an energy-efficient alternative to fully digital deep networks for real-time, power-constrained systems. However, their adoption is limited by two main challenges: the accuracy gap compared to large-scale networks during training, and discrepancies between simulated and fabricated systems that further degrade accuracy. While previous work has proposed end-to-end optimizations for specific datasets (e.g., MNIST) and optical systems, these approaches typically lack generalization across tasks and hardware designs. To address these limitations, we propose a task-agnostic and hardware-agnostic pipeline that supports image classification and segmentation across diverse optical systems. To assist optical system design before training, we estimate achievable model accuracy based on user-specified constraints such as physical size and the dataset. For training, we introduce Neural Tangent Knowledge Distillation (NTKD), which aligns optical models with electronic teacher networks, thereby narrowing the accuracy gap. After fabrication, NTKD also guides fine-tuning of the digital backend to compensate for implementation errors. Experiments on multiple datasets (e.g., MNIST, CIFAR, Carvana Masking) and hardware configurations show that our pipeline consistently improves ONN performance and enables practical deployment in both pre-fabrication simulations and physical implementations.

Trung Le, Hao Fang, Jingyuan Li et al.

Intracortical Brain-Computer Interfaces (iBCI) aim to decode behavior from neural population activity, enabling individuals with motor impairments to regain motor functions and communication abilities. A key challenge in long-term iBCI is the nonstationarity of neural recordings, where the composition and tuning profiles of the recorded populations are unstable across recording sessions. Existing methods attempt to address this issue by explicit alignment techniques; however, they rely on fixed neural identities and require test-time labels or parameter updates, limiting their generalization across sessions and imposing additional computational burden during deployment. In this work, we introduce SPINT - a Spatial Permutation-Invariant Neural Transformer framework for behavioral decoding that operates directly on unordered sets of neural units. Central to our approach is a novel context-dependent positional embedding scheme that dynamically infers unit-specific identities, enabling flexible generalization across recording sessions. SPINT supports inference on variable-size populations and allows few-shot, gradient-free adaptation using a small amount of unlabeled data from the test session. To further promote model robustness to population variability, we introduce dynamic channel dropout, a regularization method for iBCI that simulates shifts in population composition during training. We evaluate SPINT on three multi-session datasets from the FALCON Benchmark, covering continuous motor decoding tasks in human and non-human primates. SPINT demonstrates robust cross-session generalization, outperforming existing zero-shot and few-shot unsupervised baselines while eliminating the need for test-time alignment and fine-tuning. Our work contributes an initial step toward a robust and scalable neural decoding framework for long-term iBCI applications.

Caleb Zheng, Eli Shlizerman

A variety of pruning methods have been introduced for over-parameterized Recurrent Neural Networks to improve efficiency in terms of power consumption and storage utilization. These advances motivate a new paradigm, termed `hyperpruning', which seeks to identify the most suitable pruning strategy for a given network architecture and application. Unlike conventional hyperparameter search, where the optimal configuration's accuracy remains uncertain, in the context of network pruning, the accuracy of the dense model sets the target for the accuracy of the pruned one. The goal, therefore, is to discover pruned variants that match or even surpass this established accuracy. However, exhaustive search over pruning configurations is computationally expensive and lacks early performance guarantees. To address this challenge, we propose a novel Lyapunov Spectrum (LS)-based distance metric that enables early comparison between pruned and dense networks, allowing accurate prediction of post-training performance. By integrating this LS-based distance with standard hyperparameter optimization algorithms, we introduce an efficient hyperpruning framework, termed LS-based Hyperpruning (LSH). LSH reduces search time by an order of magnitude compared to conventional approaches relying on full training. Experiments on stacked LSTM and RHN architectures using the Penn Treebank dataset, and on AWD-LSTM-MoS using WikiText-2, demonstrate that under fixed training budgets and target pruning ratios, LSH consistently identifies superior pruned models. Remarkably, these pruned variants not only outperform those selected by loss-based baseline but also exceed the performance of their dense counterpart.

Jinlin Xiang, Shane Colburn, Arka Majumdar et al.

In recent years, Convolutional Neural Networks (CNNs) have enabled ubiquitous image processing applications. As such, CNNs require fast runtime (forward propagation) to process high-resolution visual streams in real time. This is still a challenging task even with state-of-the-art graphics and tensor processing units. The bottleneck in computational efficiency primarily occurs in the convolutional layers. Performing operations in the Fourier domain is a promising way to accelerate forward propagation since it transforms convolutions into elementwise multiplications, which are considerably faster to compute for large kernels. Furthermore, such computation could be implemented using an optical 4f system with orders of magnitude faster operation. However, a major challenge in using this spectral approach, as well as in an optical implementation of CNNs, is the inclusion of a nonlinearity between each convolutional layer, without which CNN performance drops dramatically. Here, we propose a Spectral CNN Linear Counterpart (SCLC) network architecture and develop a Knowledge Distillation (KD) approach to circumvent the need for a nonlinearity and successfully train such networks. While the KD approach is known in machine learning as an effective process for network pruning, we adapt the approach to transfer the knowledge from a nonlinear network (teacher) to a linear counterpart (student). We show that the KD approach can achieve performance that easily surpasses the standard linear version of a CNN and could approach the performance of the nonlinear network. Our simulations show that the possibility of increasing the resolution of the input image allows our proposed 4f optical linear network to perform more efficiently than a nonlinear network with the same accuracy on two fundamental image processing tasks: (i) object classification and (ii) semantic segmentation.

Kun Su, Xiulong Liu, Eli Shlizerman

We propose a novel system that takes as an input body movements of a musician playing a musical instrument and generates music in an unsupervised setting. Learning to generate multi-instrumental music from videos without labeling the instruments is a challenging problem. To achieve the transformation, we built a pipeline named 'Multi-instrumentalistNet' (MI Net). At its base, the pipeline learns a discrete latent representation of various instruments music from log-spectrogram using a Vector Quantized Variational Autoencoder (VQ-VAE) with multi-band residual blocks. The pipeline is then trained along with an autoregressive prior conditioned on the musician's body keypoints movements encoded by a recurrent neural network. Joint training of the prior with the body movements encoder succeeds in the disentanglement of the music into latent features indicating the musical components and the instrumental features. The latent space results in distributions that are clustered into distinct instruments from which new music can be generated. Furthermore, the VQ-VAE architecture supports detailed music generation with additional conditioning. We show that a Midi can further condition the latent space such that the pipeline will generate the exact content of the music being played by the instrument in the video. We evaluate MI Net on two datasets containing videos of 13 instruments and obtain generated music of reasonable audio quality, easily associated with the corresponding instrument, and consistent with the music audio content.

Jingyuan Li, Eli Shlizerman

Current state-of-the-art methods for skeleton-based action recognition are supervised and rely on labels. The reliance is limiting the performance due to the challenges involved in annotation and mislabeled data. Unsupervised methods have been introduced, however, they organize sequences into clusters and still require labels to associate clusters with actions. In this paper, we propose a novel approach for skeleton-based action recognition, called SESAR, that connects these approaches. SESAR leverages the information from both unlabeled data and a handful of sequences actively selected for labeling, combining unsupervised training with sparsely supervised guidance. SESAR is composed of two main components, where the first component learns a latent representation for unlabeled action sequences through an Encoder-Decoder RNN which reconstructs the sequences, and the second component performs active learning to select sequences to be labeled based on cluster and classification uncertainty. When the two components are simultaneously trained on skeleton-based action sequences, they correspond to a robust system for action recognition with only a handful of labeled samples. We evaluate our system on common datasets with multiple sequences and actions, such as NW UCLA, NTU RGB+D 60, and UWA3D. Our results outperform standalone skeleton-based supervised, unsupervised with cluster identification, and active-learning methods for action recognition when applied to sparse labeled samples, as low as 1% of the data.

Ryan Vogt, Maximilian Puelma Touzel, Eli Shlizerman et al.

Recurrent neural networks (RNNs) have been successfully applied to a variety of problems involving sequential data, but their optimization is sensitive to parameter initialization, architecture, and optimizer hyperparameters. Considering RNNs as dynamical systems, a natural way to capture stability, i.e., the growth and decay over long iterates, are the Lyapunov Exponents (LEs), which form the Lyapunov spectrum. The LEs have a bearing on stability of RNN training dynamics because forward propagation of information is related to the backward propagation of error gradients. LEs measure the asymptotic rates of expansion and contraction of nonlinear system trajectories, and generalize stability analysis to the time-varying attractors structuring the non-autonomous dynamics of data-driven RNNs. As a tool to understand and exploit stability of training dynamics, the Lyapunov spectrum fills an existing gap between prescriptive mathematical approaches of limited scope and computationally-expensive empirical approaches. To leverage this tool, we implement an efficient way to compute LEs for RNNs during training, discuss the aspects specific to standard RNN architectures driven by typical sequential datasets, and show that the Lyapunov spectrum can serve as a robust readout of training stability across hyperparameters. With this exposition-oriented contribution, we hope to draw attention to this understudied, but theoretically grounded tool for understanding training stability in RNNs.

Kun Su, Xiulong Liu, Eli Shlizerman

We present a novel system that gets as an input video frames of a musician playing the piano and generates the music for that video. Generation of music from visual cues is a challenging problem and it is not clear whether it is an attainable goal at all. Our main aim in this work is to explore the plausibility of such a transformation and to identify cues and components able to carry the association of sounds with visual events. To achieve the transformation we built a full pipeline named `\textit{Audeo}' containing three components. We first translate the video frames of the keyboard and the musician hand movements into raw mechanical musical symbolic representation Piano-Roll (Roll) for each video frame which represents the keys pressed at each time step. We then adapt the Roll to be amenable for audio synthesis by including temporal correlations. This step turns out to be critical for meaningful audio generation. As a last step, we implement Midi synthesizers to generate realistic music. \textit{Audeo} converts video to audio smoothly and clearly with only a few setup constraints. We evaluate \textit{Audeo} on `in the wild' piano performance videos and obtain that their generated music is of reasonable audio quality and can be successfully recognized with high precision by popular music identification software.

Yang Zheng, Jinlin Xiang, Kun Su et al.

We present a novel Balanced Incremental Model Agnostic Meta Learning system (BI-MAML) for learning multiple tasks. Our method implements a meta-update rule to incrementally adapt its model to new tasks without forgetting old tasks. Such a capability is not possible in current state-of-the-art MAML approaches. These methods effectively adapt to new tasks, however, suffer from 'catastrophic forgetting' phenomena, in which new tasks that are streamed into the model degrade the performance of the model on previously learned tasks. Our system performs the meta-updates with only a few-shots and can successfully accomplish them. Our key idea for achieving this is the design of balanced learning strategy for the baseline model. The strategy sets the baseline model to perform equally well on various tasks and incorporates time efficiency. The balanced learning strategy enables BI-MAML to both outperform other state-of-the-art models in terms of classification accuracy for existing tasks and also accomplish efficient adaption to similar new tasks with less required shots. We evaluate BI-MAML by conducting comparisons on two common benchmark datasets with multiple number of image classification tasks. BI-MAML performance demonstrates advantages in both accuracy and efficiency.

Jimin Kim, Eli Shlizerman

We present a novel methodology for control of neural circuits based on deep reinforcement learning. Our approach achieves aimed behavior by generating external continuous stimulation of existing neural circuits (neuromodulation control) or modulations of neural circuits architecture (connectome control). Both forms of control are challenging due to nonlinear and recurrent complexity of neural activity. To infer candidate control policies, our approach maps neural circuits and their connectome into a grid-world like setting and infers the actions needed to achieve aimed behavior. The actions are inferred by adaptation of deep Q-learning methods known for their robust performance in navigating grid-worlds. We apply our approach to the model of \textit{C. elegans} which simulates the full somatic nervous system with muscles and body. Our framework successfully infers neuropeptidic currents and synaptic architectures for control of chemotaxis. Our findings are consistent with in vivo measurements and provide additional insights into neural control of chemotaxis. We further demonstrate the generality and scalability of our methods by inferring chemotactic neural circuits from scratch.

Jingyuan Li, Eli Shlizerman

We propose a novel system for active semi-supervised feature-based action recognition. Given time sequences of features tracked during movements our system clusters the sequences into actions. Our system is based on encoder-decoder unsupervised methods shown to perform clustering by self-organization of their latent representation through the auto-regression task. These methods were tested on human action recognition benchmarks and outperformed non-feature based unsupervised methods and achieved comparable accuracy to skeleton-based supervised methods. However, such methods rely on K-Nearest Neighbours (KNN) associating sequences to actions, and general features with no annotated data would correspond to approximate clusters which could be further enhanced. Our system proposes an iterative semi-supervised method to address this challenge and to actively learn the association of clusters and actions. The method utilizes latent space embedding and clustering of the unsupervised encoder-decoder to guide the selection of sequences to be annotated in each iteration. Each iteration, the selection aims to enhance action recognition accuracy while choosing a small number of sequences for annotation. We test the approach on human skeleton-based action recognition benchmarks assuming that only annotations chosen by our method are available and on mouse movements videos recorded in lab experiments. We show that our system can boost recognition performance with only a small percentage of annotations. The system can be used as an interactive annotation tool to guide labeling efforts for 'in the wild' videos of various objects and actions to reach robust recognition.

Yang Zheng, Eli Shlizerman

Random Recurrent Neural Networks (RRNN) are the simplest recurrent networks to model and extract features from sequential data. The simplicity however comes with a price; RRNN are known to be susceptible to diminishing/exploding gradient problem when trained with gradient-descent based optimization. To enhance robustness of RRNN, alternative training approaches have been proposed. Specifically, FORCE learning approach proposed a recursive least squares alternative to train RRNN and was shown to be applicable even for the challenging task of target-learning, where the network is tasked with generating dynamic patterns with no guiding input. While FORCE training indicates that solving target-learning is possible, it appears to be effective only in a specific regime of network dynamics (edge-of-chaos). We thereby investigate whether initialization of RRNN connectivity according to a tailored distribution can guarantee robust FORCE learning. We are able to generate such distribution by inference of four generating principles constraining the spectrum of the network Jacobian to remain in stability region. This initialization along with FORCE learning provides a robust training method, i.e., Robust-FORCE (R-FORCE). We validate R-FORCE performance on various target functions for a wide range of network configurations and compare with alternative methods. Our experiments indicate that R-FORCE facilitates significantly more stable and accurate target-learning for a wide class of RRNN. Such stability becomes critical in modeling multi-dimensional sequences as we demonstrate on modeling time-series of human body joints during physical movements.

Kun Su, Xiulong Liu, Eli Shlizerman

We propose a novel system for unsupervised skeleton-based action recognition. Given inputs of body keypoints sequences obtained during various movements, our system associates the sequences with actions. Our system is based on an encoder-decoder recurrent neural network, where the encoder learns a separable feature representation within its hidden states formed by training the model to perform prediction task. We show that according to such unsupervised training the decoder and the encoder self-organize their hidden states into a feature space which clusters similar movements into the same cluster and distinct movements into distant clusters. Current state-of-the-art methods for action recognition are strongly supervised, i.e., rely on providing labels for training. Unsupervised methods have been proposed, however, they require camera and depth inputs (RGB+D) at each time step. In contrast, our system is fully unsupervised, does not require labels of actions at any stage, and can operate with body keypoints input only. Furthermore, the method can perform on various dimensions of body keypoints (2D or 3D) and include additional cues describing movements. We evaluate our system on three extensive action recognition benchmarks with different number of actions and examples. Our results outperform prior unsupervised skeleton-based methods, unsupervised RGB+D based methods on cross-view tests and while being unsupervised have similar performance to supervised skeleton-based action recognition.

Kun Su, Eli Shlizerman

Encoder-decoder recurrent neural network models (RNN Seq2Seq) have achieved great success in ubiquitous areas of computation and applications. It was shown to be successful in modeling data with both temporal and spatial dependencies for translation or prediction tasks. In this study, we propose an embedding approach to visualize and interpret the representation of data by these models. Furthermore, we show that the embedding is an effective method for unsupervised learning and can be utilized to estimate the optimality of model training. In particular, we demonstrate that embedding space projections of the decoder states of RNN Seq2Seq model trained on sequences prediction are organized in clusters capturing similarities and differences in the dynamics of these sequences. Such performance corresponds to an unsupervised clustering of any spatio-temporal features and can be employed for time-dependent problems such as temporal segmentation, clustering of dynamic activity, self-supervised classification, action recognition, failure prediction, etc. We test and demonstrate the application of the embedding methodology to time-sequences of 3D human body poses. We show that the methodology provides a high-quality unsupervised categorization of movements.

Shane Colburn, Yi Chu, Eli Shlizerman et al.

The parallelism of optics and the miniaturization of optical components using nanophotonic structures, such as metasurfaces present a compelling alternative to electronic implementations of convolutional neural networks. The lack of a low-power optical nonlinearity, however, requires slow and energy-inefficient conversions between the electronic and optical domains. Here, we design an architecture which utilizes a single electrical to optical conversion by designing a free-space optical frontend unit that implements the linear operations of the first layer with the subsequent layers realized electronically. Speed and power analysis of the architecture indicates that the hybrid photonic-electronic architecture outperforms sole electronic architecture for large image sizes and kernels. Benchmarking of the photonic-electronic architecture on a modified version of AlexNet achieves a classification accuracy of 87% on images from the Kaggle Cats and Dogs challenge database.

Eli Shlizerman, Lucio M. Dery, Hayden Schoen et al.

We present a method that gets as input an audio of violin or piano playing, and outputs a video of skeleton predictions which are further used to animate an avatar. The key idea is to create an animation of an avatar that moves their hands similarly to how a pianist or violinist would do, just from audio. Aiming for a fully detailed correct arms and fingers motion is a goal, however, it's not clear if body movement can be predicted from music at all. In this paper, we present the first result that shows that natural body dynamics can be predicted at all. We built an LSTM network that is trained on violin and piano recital videos uploaded to the Internet. The predicted points are applied onto a rigged avatar to create the animation.