arXiv subjects

Alireza Morsali

Publications and source records attributed to Alireza Morsali.

7 recordsLinked to original sources

arXiv2026

Coverage Matters: MarginMerge for Compressing Multi-Vector Visual Document Retrievers

Multi-vector visual document retrievers such as ColPali and ColQwen achieve strong retrieval by storing fine-grained patch embeddings, but this produces large indexes and costly late-interaction scoring. We argue that effective compression should preserve query-relevant coverage, meaning the diverse document regions that may become the strongest MaxSim match across queries, rather than selecting patches independently by salience. This view also explains why dense rendered pages are easier to compress than natural images. We introduce MarginMerge, a compression method for frozen multi-vector retrievers. It selects coverage-aware anchors, clusters document patches, and uses a lightweight shared network to synthesize one representative per cluster. Compression is performed once during indexing, while retrieval keeps the standard MaxSim interface. Across six datasets on both ColQwen2.5 and ColPali, MarginMerge achieves the highest matched query-agnostic average at 5% and 10% vector retention. Compared with the uncompressed index using the same backbone, it preserves between 97% and 99% of average nDCG@5 while reducing stored document vectors by between 90% and 95%. At 5% retention, it also reduces ranking flips relative to geometric merging on all six ColQwen2.5 datasets by approximately 41% on average. The same model transfers to unseen datasets and retention ratios without retraining.

cs.IR↗

arXiv2026

A Unified Theory of Sinusoidal Activation Families for Implicit Neural Representations

Implicit Neural Representations (INRs) model continuous signals with compact neural networks and have become a standard tool in vision, graphics, and signal processing. A central challenge is accurately capturing fine detail without heavy hand-crafted encodings or brittle training heuristics. Across the literature, periodic activations have emerged as a compelling remedy: from SIREN, which uses a single sinusoid with a fixed global frequency, to more recent architectures employing multiple sinusoids and, in some cases, trainable frequencies and phases. We study this family of sinusoidal activations and develop a principled theoretical and practical framework for trainable sinusoidal activations in INRs. Concretely, we instantiate this framework with Sinusoidal Trainable Activation Functions (STAF), a Fourier-like activation whose amplitudes, frequencies, and phases are learned. Our analysis (i) establishes a Kronecker-equivalence construction that expresses trainable sinusoidal activations with standard sine networks and quantifies expressive growth, (ii) characterizes how the Neural Tangent Kernel (NTK) spectrum changes under trainable sinusoidal parameterization, and (iii) provides an initialization that yields standard normal post-activations without asymptotic central limit theorem (CLT) arguments. Empirically, on images, audio, shapes, inverse problems (super-resolution, denoising) and NeRF, STAF is competitive and often stronger on distortion-oriented reconstruction metrics such as PSNR/SSIM across the evaluated INR tasks, with favorable parameter efficiency under layer-wise sharing. While periodic activations can alleviate practical manifestations of spectral bias, our results indicate they do not eliminate it; instead, trainable sinusoids can improve the observed capacity-optimization trade-off in the evaluated settings.

cs.CV↗

arXiv2023

DiffGANPaint: Fast Inpainting Using Denoising Diffusion GANs

Free-form image inpainting is the task of reconstructing parts of an image specified by an arbitrary binary mask. In this task, it is typically desired to generalize model capabilities to unseen mask types, rather than learning certain mask distributions. Capitalizing on the advances in diffusion models, in this paper, we propose a Denoising Diffusion Probabilistic Model (DDPM) based model capable of filling missing pixels fast as it models the backward diffusion process using the generator of a generative adversarial network (GAN) network to reduce sampling cost in diffusion models. Experiments on general-purpose image inpainting datasets verify that our approach performs superior or on par with most contemporary works.

cs.CV↗

arXiv2022

Ensemble Neural Representation Networks

Implicit Neural Representation (INR) has recently attracted considerable attention for storing various types of signals in continuous forms. The existing INR networks require lengthy training processes and high-performance computational resources. In this paper, we propose a novel sub-optimal ensemble architecture for INR that resolves the aforementioned problems. In this architecture, the representation task is divided into several sub-tasks done by independent sub-networks. We show that the performance of the proposed ensemble INR architecture may decrease if the dimensions of sub-networks increase. Hence, it is vital to suggest an optimization algorithm to find the sub-optimal structure of the ensemble network, which is done in this paper. According to the simulation results, the proposed architecture not only has significantly fewer floating-point operations (FLOPs) and less training time, but it also has better performance in terms of Peak Signal to Noise Ratio (PSNR) compared to those of its counterparts.

cs.LG↗

arXiv2022

Multi-Head ReLU Implicit Neural Representation Networks

In this paper, a novel multi-head multi-layer perceptron (MLP) structure is presented for implicit neural representation (INR). Since conventional rectified linear unit (ReLU) networks are shown to exhibit spectral bias towards learning low-frequency features of the signal, we aim at mitigating this defect by taking advantage of the local structure of the signals. To be more specific, an MLP is used to capture the global features of the underlying generator function of the desired signal. Then, several heads are utilized to reconstruct disjoint local features of the signal, and to reduce the computational complexity, sparse layers are deployed for attaching heads to the body. Through various experiments, we show that the proposed model does not suffer from the special bias of conventional ReLU networks and has superior generalization capabilities. Finally, simulation results confirm that the proposed multi-head structure outperforms existing INR methods with considerably less computational cost.

cs.LG↗

arXiv2021

Light-SERNet: A lightweight fully convolutional neural network for speech emotion recognition

Detecting emotions directly from a speech signal plays an important role in effective human-computer interactions. Existing speech emotion recognition models require massive computational and storage resources, making them hard to implement concurrently with other machine-interactive tasks in embedded systems. In this paper, we propose an efficient and lightweight fully convolutional neural network for speech emotion recognition in systems with limited hardware resources. In the proposed FCNN model, various feature maps are extracted via three parallel paths with different filter sizes. This helps deep convolution blocks to extract high-level features, while ensuring sufficient separability. The extracted features are used to classify the emotion of the input speech segment. While our model has a smaller size than that of the state-of-the-art models, it achieves higher performance on the IEMOCAP and EMO-DB datasets.

eess.AS↗

arXiv2021

Deep Learning Framework for Hybrid Analog-Digital Signal Processing in mmWave Massive-MIMO Systems

Hybrid analog-digital signal processing (HSP) is an enabling technology to harvest the potential of millimeter-wave (mmWave) massive-MIMO communications. In this paper, we present a general deep learning (DL) framework for efficient design and implementation of HSP-based massive-MIMO systems. Exploiting the fact that any complex matrix can be written as a scaled sum of two matrices with unit-modulus entries, a novel analog deep neural network (ADNN) structure is first developed which can be implemented with common radio frequency (RF) components. This structure is then embedded into an extended hybrid analog-digital deep neural network (HDNN) architecture which facilitates the implementation of mmWave massive-MIMO systems while improving their performance. In particular, the proposed HDNN architecture enables HSP-based massive-MIMO transceivers to approximate any desired transmitter and receiver mapping with arbitrary precision. To demonstrate the capabilities of the proposed DL framework, we present a new HDNN-based beamformer design that can achieve the same performance as fully-digital beamforming, with reduced number of RF chains. Finally, simulation results are presented confirming the superiority of the proposed HDNN design over existing hybrid beamforming schemes.

eess.SP↗