arXiv subjects

Ali Momeni

Publications and source records attributed to Ali Momeni.

At least 19 recordsLinked to original sources

arXiv2026

Multi-channel Optical Vision Model

Spatial multiplexing is one of the natural strengths of optics, yet in optical neural networks, it is often used mainly as parallel throughput. Here, we show that spatial multiplexing in an optical neural network can be used not only to process multiple inputs in parallel, but also to define a trainable representational coordinate of the model. In three implemented scenarios, parallel-input processing, class-code readout and channel-mixed feature interaction, spatial channels act as independent learners, structured code dimensions, and interacting feature groups. The programmable free-space optical processor is trained through an online physical-forward/surrogate-backward scheme, where measured optical outputs define the forward pass while a differentiable surrogate estimates gradients and is continually fine-tuned during training from newly acquired optical data. We demonstrate these channel roles in image classification and regression tasks using multi-layer architectures with more than one million trainable optical phase parameters. We further implement a hybrid optical-electronic vision-language model, in which the optical neural network provides visual tokens to a digital transformer decoder for controlled image-captioning tasks. These results establish spatially multiplexed optical channels as a programmable feature and readout space for hybrid optical vision models.

physics.optics↗

arXiv2026

Towards Engineering Material Neural Networks

Structures that capture functionality in the form of animate or intelligent machines have the potential to transform modern engineering applications. Animation and embedded intelligence are typically realised by integrating advanced capabilities such as reversibility, adaptive responses and learning directly into the materials themselves. Currently, the majority of adaptive material systems rely on predefined adaptive designs combined with in-service, electronics-based computing to dynamically modify the structural behaviour. However, structural configurations with interconnected adaptable nodes are able to approximate continuous functions, providing new possibilities and opportunities than classical metamaterials and computational materials. We discuss here the potential to design load-bearing engineering materials with trainable physical parameters and neural network-inspired morphologies, embedding intelligence directly into their structure, a concept we define as Engineering Material Neural Networks (EMNNs) as a subcategory of Physical Neural Networks. In this perspective, we first establish the foundational concept of EMNNs; we then detail the mechanical and multifunctional properties required for such structural configurations. Finally, we evaluate existing and emerging engineering materials that hold promise for enabling this innovative approach. Key material candidates for realising EMNNs include composites, architected, biological and engineering living materials. We also outline future directions in materials science and structural engineering for developing EMNNs.

cond-mat.mtrl-sci↗

arXiv2026

Expressivity of Programmable-Metasurface-Based Physical Neural Networks: Encoding Non-Linearity, Structural Non-Linearity, and Depth

Wave-based signal processing conventionally encodes input data into the input wavefront, making it challenging to implement non-linear operations. Programmable wave systems enable an alternative approach: encoding the input data into the scattering properties of tunable components. With such structural input encoding, two potentially non-linear mappings are involved: first, from the input data to the tunable components' scattering characteristics, and, second, from these scattering characteristics to the output wavefront. In this paper, we systematically examine the expressivity of a wave-based physical neural network (WPNN) with structural input encoding. Our analysis is based on a physics-consistent multiport-network model of a compact D-band rich-scattering cavity parametrized by a 100-element programmable metasurface. We separately control encoding non-linearity, structural non-linearity, and network depth in order to examine their interplay, considering a controlled scalar regression task. With phase encoding and strong inter-element mutual coupling (MC), both aforementioned mappings are strongly non-linear and the WPNN performs very well even with a single layer. We further observe that additional layers can partially compensate for weak inter-element MC. In addition, we demonstrate that WPNN depth can improve expressivity even when it is not associated with an increase in trainable weights. Altogether, our results provide a physics-consistent picture of how encoding choice, MC strength, and depth jointly govern the expressive power of PM-based WPNNs, informing design choices for future experimental implementations of WPNNs.

eess.SP↗

arXiv2026

Dynamic Hierarchical $j$-Tree Decomposition and Its Applications

We develop a new algorithmic framework for designing approximation algorithms for cut-based optimization problems on capacitated undirected graphs that undergo edge insertions and deletions. Specifically, our framework dynamically maintains a variant of the hierarchical $j$-tree decomposition of [Madry FOCS'10], achieving a poly-logarithmic approximation factor to the graph's cut structure and supporting edge updates in $O(n^{\epsilon})$ amortized update time, for any arbitrarily small constant $\epsilon \in (0,1)$. Consequently, we obtain new trade-offs between approximation and update/query time for fundamental cut-based optimization problems in the fully dynamic setting, including all-pairs minimum cuts, sparsest cut, multi-way cut, and multi-cut. For the last three problems, these trade-offs give the first fully-dynamic algorithms achieving poly-logarithmic approximation in sub-linear time per operation. The main technical ingredient behind our dynamic hierarchy is a dynamic cut-sparsifier algorithm that can handle vertex splits with low recourse. This is achieved by white-boxing the dynamic cut sparsifier construction of [Abraham et al. FOCS'16], based on forest packing, together with new structural insights about the maintenance of these forests under vertex splits. Given the versatility of cut sparsification in both the static and dynamic graph algorithms literature, we believe this construction may be of independent interest.

cs.DS↗

arXiv2025

Fully Dynamic Spectral Sparsification for Directed Hypergraphs

There has been a surge of interest in spectral hypergraph sparsification, a natural generalization of spectral sparsification for graphs. In this paper, we present a simple fully dynamic algorithm for maintaining spectral hypergraph sparsifiers of \textit{directed} hypergraphs. Our algorithm achieves a near-optimal size of $O(n^2 / \varepsilon ^2 \log ^7 m)$ and amortized update time of $O(r^2 \log ^3 m)$, where $n$ is the number of vertices, and $m$ and $r$ respectively upper bound the number of hyperedges and the rank of the hypergraph at any time. We also extend our approach to the parallel batch-dynamic setting, where a batch of any $k$ hyperedge insertions or deletions can be processed with $O(kr^2 \log ^3 m)$ amortized work and $O(\log ^2 m)$ depth. This constitutes the first spectral-based sparsification algorithm in this setting.

cs.DS↗

arXiv2025

Fully Dynamic Algorithms for Transitive Reduction

Given a directed graph $G$, a transitive reduction $G^t$ of $G$ (first studied by Aho, Garey, Ullman [SICOMP `72]) is a minimal subgraph of $G$ that preserves the reachability relation between every two vertices in $G$. In this paper, we study the computational complexity of transitive reduction in the dynamic setting. We obtain the first fully dynamic algorithms for maintaining a transitive reduction of a general directed graph undergoing updates such as edge insertions or deletions. Our first algorithm achieves $O(m+n \log n)$ amortized update time, which is near-optimal for sparse directed graphs, and can even support extended update operations such as inserting a set of edges all incident to the same vertex, or deleting an arbitrary set of edges. Our second algorithm relies on fast matrix multiplication and achieves $O(m+ n^{1.585})$ \emph{worst-case} update time.

cs.DS↗

arXiv2025

Fully Dynamic Spectral Sparsification of Hypergraphs

Spectral hypergraph sparsification, a natural generalization of the well-studied spectral sparsification notion on graphs, has been the subject of intensive research in recent years. In this work, we consider spectral hypergraph sparsification in the dynamic setting, where the goal is to maintain a spectral sparsifier of an undirected, weighted hypergraph subject to a sequence of hyperedge insertions and deletions. For any $0 < \varepsilon \leq 1$, we give the first fully dynamic algorithm for maintaining an $ (1 \pm \varepsilon) $-spectral hypergraph sparsifier of size $ n r^3 \operatorname{poly}\left( \log n, \varepsilon ^{-1} \right) $ with amortized update time $ r^4 \operatorname{poly}\left( \log n, \varepsilon ^{-1} \right) $, where $n$ is the number of vertices of the underlying hypergraph and $r$ is an upper-bound on the rank of hyperedges. Our key contribution is to show that the spanner-based sparsification algorithm of Koutis and Xu (2016) admits a dynamic implementation in the hypergraph setting, thereby extending the dynamic spectral sparsification framework for ordinary graphs by Abraham et al. (2016).

cs.DS↗

arXiv2025

Locality-aware Surrogates for Gradient-based Black-box Optimization

In physics and engineering, many processes are modeled using non-differentiable black-box simulators, making the optimization of such functions particularly challenging. To address such cases, inspired by the Gradient Theorem, we propose locality-aware surrogate models for active model-based black-box optimization. We first establish a theoretical connection between gradient alignment and the minimization of a Gradient Path Integral Equation (GradPIE) loss, which enforces consistency of the surrogate's gradients in local regions of the design space. Leveraging this theoretical insight, we develop a scalable training algorithm that minimizes the GradPIE loss, enabling both offline and online learning while maintaining computational efficiency. We evaluate our approach on three real-world tasks - spanning automated in silico experiments such as coupled nonlinear oscillators, analog circuits, and optical systems - and demonstrate consistent improvements in optimization efficiency under limited query budgets. Our results offer dependable solutions for both offline and online optimization tasks where reliable gradient estimation is needed.

cs.LG↗

arXiv2024

GraCo -- A Graph Composer for Integrated Circuits

Designing integrated circuits involves substantial complexity, posing challenges in revealing its potential applications - from custom digital cells to analog circuits. Despite extensive research over the past decades in building versatile and automated frameworks, there remains open room to explore more computationally efficient AI-based solutions. This paper introduces the graph composer GraCo, a novel method for synthesizing integrated circuits using reinforcement learning (RL). GraCo learns to construct a graph step-by-step, which is then converted into a netlist and simulated with SPICE. We demonstrate that GraCo is highly configurable, enabling the incorporation of prior design knowledge into the framework. We formalize how this prior knowledge can be utilized and, in particular, show that applying consistency checks enhances the efficiency of the sampling process. To evaluate its performance, we compare GraCo to a random baseline, which is known to perform well for smaller design space problems. We demonstrate that GraCo can discover circuits for tasks such as generating standard cells, including the inverter and the two-input NAND (NAND2) gate. Compared to a random baseline, GraCo requires 5x fewer sampling steps to design an inverter and successfully synthesizes a NAND2 gate that is 2.5x faster.

cs.LG↗

arXiv2024

Schemato -- An LLM for Netlist-to-Schematic Conversion

Machine learning models are advancing circuit design, particularly in analog circuits. They typically generate netlists that lack human interpretability. This is a problem as human designers heavily rely on the interpretability of circuit diagrams or schematics to intuitively understand, troubleshoot, and develop designs. Hence, to integrate domain knowledge effectively, it is crucial to translate ML-generated netlists into interpretable schematics quickly and accurately. We propose Schemato, a large language model (LLM) for netlist-to-schematic conversion. In particular, we consider our approach in converting netlists to .asc files, text-based schematic description used in LTSpice. Experiments on our circuit dataset show that Schemato achieves up to 76% compilation success rate, surpassing 63% scored by the state-of-the-art LLMs. Furthermore, our experiments show that Schemato generates schematics with an average graph edit distance score and mean structural similarity index measure, scaled by the compilation success rate that are 1.8x and 4.3x higher than the best performing LLMs respectively, demonstrating its ability to generate schematics that are more accurately connected and are closer to the reference human design.

cs.LG↗

arXiv2024

Training of Physical Neural Networks

Physical neural networks (PNNs) are a class of neural-like networks that leverage the properties of physical systems to perform computation. While PNNs are so far a niche research area with small-scale laboratory demonstrations, they are arguably one of the most underappreciated important opportunities in modern AI. Could we train AI models 1000x larger than current ones? Could we do this and also have them perform inference locally and privately on edge devices, such as smartphones or sensors? Research over the past few years has shown that the answer to all these questions is likely "yes, with enough research": PNNs could one day radically change what is possible and practical for AI systems. To do this will however require rethinking both how AI models work, and how they are trained - primarily by considering the problems through the constraints of the underlying hardware physics. To train PNNs at large scale, many methods including backpropagation-based and backpropagation-free approaches are now being explored. These methods have various trade-offs, and so far no method has been shown to scale to the same scale and performance as the backpropagation algorithm widely used in deep learning today. However, this is rapidly changing, and a diverse ecosystem of training techniques provides clues for how PNNs may one day be utilized to create both more efficient realizations of current-scale AI models, and to enable unprecedented-scale models.

physics.app-ph↗

arXiv2024

Finding d-Cuts in Graphs of Bounded Diameter, Graphs of Bounded Radius and H-Free Graphs

The d-Cut problem is to decide if a graph has an edge cut such that each vertex has at most d neighbours at the opposite side of the cut. If $d=1$, we obtain the intensively studied Matching Cut problem. The d-Cut problem has been studied as well, but a systematic study for special graph classes was lacking. We initiate such a study and consider classes of bounded diameter, bounded radius and $H$-free graphs. We prove that for all $d\geq 2$, d-Cut is polynomial-time solvable for graphs of diameter 2, $(P_3+P_4)$-free graphs and $P_5$-free graphs. These results extend known results for $d=1$. However, we also prove several NP-hardness results for d-Cut that contrast known polynomial-time results for $d=1$. Our results lead to full dichotomies for bounded diameter and bounded radius and to almost-complete dichotomies for H-free graphs.

math.CO↗

arXiv2023

Backpropagation-free Training of Deep Physical Neural Networks

Recent years have witnessed the outstanding success of deep learning in various fields such as vision and natural language processing. This success is largely indebted to the massive size of deep learning models that is expected to increase unceasingly. This growth of the deep learning models is accompanied by issues related to their considerable energy consumption, both during the training and inference phases, as well as their scalability. Although a number of work based on unconventional physical systems have been proposed which addresses the issue of energy efficiency in the inference phase, efficient training of deep learning models has remained unaddressed. So far, training of digital deep learning models mainly relies on backpropagation, which is not suitable for physical implementation as it requires perfect knowledge of the computation performed in the so-called forward pass of the neural network. Here, we tackle this issue by proposing a simple deep neural network architecture augmented by a biologically plausible learning algorithm, referred to as "model-free forward-forward training". The proposed architecture enables training deep physical neural networks consisting of layers of physical nonlinear systems, without requiring detailed knowledge of the nonlinear physical layers' properties. We show that our method outperforms state-of-the-art hardware-aware training methods by improving training speed, decreasing digital computations, and reducing power consumption in physical systems. We demonstrate the adaptability of the proposed method, even in systems exposed to dynamic or unpredictable external perturbations. To showcase the universality of our approach, we train diverse wave-based physical neural networks that vary in the underlying wave phenomenon and the type of non-linearity they use, to perform vowel and image classification tasks experimentally.

cs.LG↗

arXiv2023

Physics-inspired Neuroacoustic Computing Based on Tunable Nonlinear Multiple-scattering

Waves, such as light and sound, inherently bounce and mix due to multiple scattering induced by the complex material objects that surround us. This scattering process severely scrambles the information carried by waves, challenging conventional communication systems, sensing paradigms, and wave-based computing schemes. Here, we show that instead of being a hindrance, multiple scattering can be beneficial to enable and enhance analog nonlinear information mapping, allowing for the direct physical implementation of computational paradigms such as reservoir computing and extreme learning machines. We propose a physics-inspired version of such computational architectures for speech and vowel recognition that operate directly in the native domain of the input signal, namely on real-sounds, without any digital pre-processing or encoding conversion and backpropagation training computation. We first implement it in a proof-of-concept prototype, a nonlinear chaotic acoustic cavity containing multiple tunable and power-efficient nonlinear meta-scatterers. We prove the efficiency of the acoustic-based computing system for vowel recognition tasks with high testing classification accuracy (91.4%). Finally, we demonstrate the high performance of vowel recognition in the natural environment of a reverberation room. Our results open the way for efficient acoustic learning machines that operate directly on the input sound, and leverage physics to enable Natural Language Processing (NLP).

cs.SD↗

arXiv2022

Near-Field Optical MIMO Communication with Polarization-dependent Metasurfaces

The ability to control waves at the nanoscale has attracted considerable attention to ultrathin metasurface lenses (metalenses) in optical imaging and encryption systems. We propose an approach to active tuning metasurfaces by integrating an ultrathin layer of indium-tin-oxide (ITO) into a unit cell as an electro-optically tunable material. A proposed design features two orthogonal wings that can independently manipulate waves with corresponding orthogonal polarizations. The charge carrier concentration in the ITO accumulation layer is altered by modulating the applied bias voltage. This bias voltage generates phase variations at terahertz frequencies for the reflected transverse electric and transverse magnetic polarized waves. It is possible to move both focal points of a metalens without any physical movement by varying the bias voltage. In addition, this paper explores the application of virtually moving metalens for a novel multiple-input and multiple-output (MIMO) communication architecture. We demonstrate a communication system based on single-point binary data communication and hexadecimal orbital angular momentum (OAM) data communication. Then, orthogonal channels can be used for MIMO communication with high capacity. The proposed design paves the way for high-speed communications as well as polarization-controlled molecular imaging systems.

physics.optics↗

arXiv2022

The Game of Cops and Robber on (Claw, Even-hole)-free Graphs

In this paper, we study the game of cops and robber on the class of graphs with no even hole (induced cycle of even length) and claw (a star with three leaves). The cop number of a graph $G$ is defined as the minimum number of cops needed to capture the robber. Here, we prove that the cop number of all claw-free even-hole-free graphs is at most two and, in addition, the capture time is at most $2n$ rounds, where $n$ is the number of vertices of the graph. Moreover, our results can be viewed as a first step towards studying the structure of claw-free even-hole-free graphs.

math.CO↗

arXiv2021

A single metagrating metastructure for wave-based parallel analog computing

Wave-based signal processing has witnessed a significant expansion of interest in a variety of science and engineering disciplines, as it provides new opportunities for achieving high-speed and low-power operations. Although flat optics desires integrable components to perform multiple missions, yet, the current wave-based analog computers can engineer only the spatial content of the input signal where the processed signal obeys the traditional version of Snell's law. In this paper, we propose a multi-functional metagrating to modulate both spatial and angular properties of the input signal whereby both symmetric and asymmetric optical transfer functions are realized using high-order space harmonics. The performance of the designed compound metallic grating is validated through several investigations where closed-form expressions are suggested to extract the phase and amplitude information of the diffractive modes. Several illustrative examples are demonstrated to show that the proposed metagrating allows for simultaneous parallel analog computing tasks such as first- and second-order spatial differentiation through a single multi-channel structured surface. It is anticipated that the designed platform brings a new twist to the field of optical signal processing and opens up large perspectives for simple integrated image processing systems.

physics.optics↗

arXiv2021

Wave-based extreme deep learning based on non-linear time-Floquet entanglement

Wave-based analog signal processing holds the promise of extremely fast, on-the-fly, power-efficient data processing, occurring as a wave propagates through an artificially engineered medium. Yet, due to the fundamentally weak non-linearities of traditional wave materials, such analog processors have been so far largely confined to simple linear projections such as image edge detection or matrix multiplications. Complex neuromorphic computing tasks, which inherently require strong non-linearities, have so far remained out-of-reach of wave-based solutions, with a few attempts that implemented non-linearities on the digital front, or used weak and inflexible non-linear sensors, restraining the learning performance. Here, we tackle this issue by demonstrating the relevance of Time-Floquet physics to induce a strong non-linear entanglement between signal inputs at different frequencies, enabling a power-efficient and versatile wave platform for analog extreme deep learning involving a single, uniformly modulated dielectric layer and a scattering medium. We prove the efficiency of the method for extreme learning machines and reservoir computing to solve a range of challenging learning tasks, from forecasting chaotic time series to the simultaneous classification of distinct datasets. Our results open the way for wave-based machine learning with high energy efficiency, speed, and scalability.

cs.ET↗