Alireza Mousavi

Abdelrahman Abouelenin, Atabak Ashfaq, Adam Atkinson et al. · microsoft-research

We introduce Phi-4-Mini and Phi-4-Multimodal, compact yet highly capable language and multimodal models. Phi-4-Mini is a 3.8-billion-parameter language model trained on high-quality web and synthetic data, significantly outperforming recent open-source models of similar size and matching the performance of models twice its size on math and coding tasks requiring complex reasoning. This achievement is driven by a carefully curated synthetic data recipe emphasizing high-quality math and coding datasets. Compared to its predecessor, Phi-3.5-Mini, Phi-4-Mini features an expanded vocabulary size of 200K tokens to better support multilingual applications, as well as group query attention for more efficient long-sequence generation. Phi-4-Multimodal is a multimodal model that integrates text, vision, and speech/audio input modalities into a single model. Its novel modality extension approach leverages LoRA adapters and modality-specific routers to allow multiple inference modes combining various modalities without interference. For example, it now ranks first in the OpenASR leaderboard to date, although the LoRA component of the speech/audio modality has just 460 million parameters. Phi-4-Multimodal supports scenarios involving (vision + language), (vision + speech), and (speech/audio) inputs, outperforming larger vision-language and speech-language models on a wide range of tasks. Additionally, we experiment to further train Phi-4-Mini to enhance its reasoning capabilities. Despite its compact 3.8-billion-parameter size, this experimental version achieves reasoning performance on par with or surpassing significantly larger models, including DeepSeek-R1-Distill-Qwen-7B and DeepSeek-R1-Distill-Llama-8B.

Ali Mousavi, Xin Zhan, He Bai et al. · apple-ml

Datasets that pair Knowledge Graphs (KG) and text together (KG-T) can be used to train forward and reverse neural models that generate text from KG and vice versa. However models trained on datasets where KG and text pairs are not equivalent can suffer from more hallucination and poorer recall. In this paper, we verify this empirically by generating datasets with different levels of noise and find that noisier datasets do indeed lead to more hallucination. We argue that the ability of forward and reverse models trained on a dataset to cyclically regenerate source KG or text is a proxy for the equivalence between the KG and the text in the dataset. Using cyclic evaluation we find that manually created WebNLG is much better than automatically created TeKGen and T-REx. Guided by these observations, we construct a new, improved dataset called LAGRANGE using heuristics meant to improve equivalence between KG and text and show the impact of each of the heuristics on cyclic evaluation. We also construct two synthetic datasets using large language models (LLMs), and observe that these are conducive to models that perform significantly well on cyclic generation of text, but less so on cyclic generation of KGs, probably because of a lack of a consistent underlying ontology.

Jason Mohoney, Anil Pacaci, Shihabur Rahman Chowdhury et al.

There is an increasing adoption of machine learning for encoding data into vectors to serve online recommendation and search use cases. As a result, recent data management systems propose augmenting query processing with online vector similarity search. In this work, we explore vector similarity search in the context of Knowledge Graphs (KGs). Motivated by the tasks of finding related KG queries and entities for past KG query workloads, we focus on hybrid vector similarity search (hybrid queries for short) where part of the query corresponds to vector similarity search and part of the query corresponds to predicates over relational attributes associated with the underlying data vectors. For example, given past KG queries for a song entity, we want to construct new queries for new song entities whose vector representations are close to the vector representation of the entity in the past KG query. But entities in a KG also have non-vector attributes such as a song associated with an artist, a genre, and a release date. Therefore, suggested entities must also satisfy query predicates over non-vector attributes beyond a vector-based similarity predicate. While these tasks are central to KGs, our contributions are generally applicable to hybrid queries. In contrast to prior works that optimize online queries, we focus on enabling efficient batch processing of past hybrid query workloads. We present our system, HQI, for high-throughput batch processing of hybrid queries. We introduce a workload-aware vector data partitioning scheme to tailor the vector index layout to the given workload and describe a multi-query optimization technique to reduce the overhead of vector similarity computations. We evaluate our methods on industrial workloads and demonstrate that HQI yields a 31x improvement in throughput for finding related KG queries compared to existing hybrid query processing approaches.

Ronak Pradeep, Daniel Lee, Ali Mousavi et al.

The rapid advancement of Large Language Models (LLMs) and conversational assistants necessitates dynamic, scalable, and configurable conversational datasets for training and evaluation. These datasets must accommodate diverse user interaction modes, including text and voice, each presenting unique modeling challenges. Knowledge Graphs (KGs), with their structured and evolving nature, offer an ideal foundation for current and precise knowledge. Although human-curated KG-based conversational datasets exist, they struggle to keep pace with the rapidly changing user information needs. We present ConvKGYarn, a scalable method for generating up-to-date and configurable conversational KGQA datasets. Qualitative psychometric analyses confirm our method can generate high-quality datasets rivaling a popular conversational KGQA dataset while offering it at scale and covering a wide range of human-interaction configurations. We showcase its utility by testing LLMs on diverse conversations - exploring model behavior on conversational KGQA sets with different configurations grounded in the same KG fact set. Our results highlight the ability of ConvKGYarn to improve KGQA foundations and evaluate parametric knowledge of LLMs, thus offering a robust solution to the constantly evolving landscape of conversational assistants.

Christopher A. Metzler, Ali Mousavi, Reinhard Heckel et al.

Learning from unlabeled and noisy data is one of the grand challenges of machine learning. As such, it has seen a flurry of research with new ideas proposed continuously. In this work, we revisit a classical idea: Stein's Unbiased Risk Estimator (SURE). We show that, in the context of image recovery, SURE and its generalizations can be used to train convolutional neural networks (CNNs) for a range of image denoising and recovery problems without any ground truth data. Specifically, our goal is to reconstruct an image $x$ from a noisy linear transformation (measurement) of the image. We consider two scenarios: one where no additional data is available and one where we have measurements of other images that are drawn from the same noisy distribution as $x$, but have no access to the clean images. Such is the case, for instance, in the context of medical imaging, microscopy, and astronomy, where noise-less ground truth data is rarely available. We show that in this situation, SURE can be used to estimate the mean-squared-error loss associated with an estimate of $x$. Using this estimate of the loss, we train networks to perform denoising and compressed sensing recovery. In addition, we also use the SURE framework to partially explain and improve upon an intriguing results presented by Ulyanov et al. in "Deep Image Prior": that a network initialized with random weights and fit to a single noisy image can effectively denoise that image. Public implementations of the networks and methods described in this paper can be found at https://github.com/ricedsp/D-AMP_Toolbox.

Maliheh Toozandehjani, Ali Mousavi, Reza Taheri

The aim of image-based virtual try-on is to generate realistic images of individuals wearing target garments, ensuring that the pose, body shape and characteristics of the target garment are accurately preserved. Existing methods often fail to reproduce the fine details of target garments effectively and lack generalizability to new scenarios. In the proposed method, the person's initial garment is completely removed. Subsequently, a precise warping is performed using the predicted keypoints to fully align the target garment with the body structure and pose of the individual. Based on the warped garment, a body segmentation map is more accurately predicted. Then, using an alignment-aware segment normalization, the misaligned areas between the warped garment and the predicted garment region in the segmentation map are removed. Finally, the generator produces the final image with high visual quality, reconstructing the precise characteristics of the target garment, including its overall shape and texture. This approach emphasizes preserving garment characteristics and improving adaptability to various poses, providing better generalization for diverse applications.

Ali Mousavi, Lihong Li, Qiang Liu et al.

Off-policy estimation for long-horizon problems is important in many real-life applications such as healthcare and robotics, where high-fidelity simulators may not be available and on-policy evaluation is expensive or impossible. Recently, \cite{liu18breaking} proposed an approach that avoids the \emph{curse of horizon} suffered by typical importance-sampling-based methods. While showing promising results, this approach is limited in practice as it requires data be drawn from the \emph{stationary distribution} of a \emph{known} behavior policy. In this work, we propose a novel approach that eliminates such limitations. In particular, we formulate the problem as solving for the fixed point of a certain operator. Using tools from Reproducing Kernel Hilbert Spaces (RKHSs), we develop a new estimator that computes importance ratios of stationary distributions, without knowledge of how the off-policy data are collected. We analyze its asymptotic consistency and finite-sample generalization. Experiments on benchmarks verify the effectiveness of our approach.

Ali Mousavi, Gautam Dasarathy, Richard G. Baraniuk

In this paper we develop a novel computational sensing framework for sensing and recovering structured signals. When trained on a set of representative signals, our framework learns to take undersampled measurements and recover signals from them using a deep convolutional neural network. In other words, it learns a transformation from the original signals to a near-optimal number of undersampled measurements and the inverse transformation from measurements to signals. This is in contrast to traditional compressive sensing (CS) systems that use random linear measurements and convex optimization or iterative algorithms for signal recovery. We compare our new framework with $\ell_1$-minimization from the phase transition point of view and demonstrate that it outperforms $\ell_1$-minimization in the regions of phase transition plot where $\ell_1$-minimization cannot recover the exact solution. In addition, we experimentally demonstrate how learning measurements enhances the overall recovery performance, speeds up training of recovery framework, and leads to having fewer parameters to learn.

Christopher A. Metzler, Ali Mousavi, Richard G. Baraniuk

Compressive image recovery is a challenging problem that requires fast and accurate algorithms. Recently, neural networks have been applied to this problem with promising results. By exploiting massively parallel GPU processing architectures and oodles of training data, they can run orders of magnitude faster than existing techniques. However, these methods are largely unprincipled black boxes that are difficult to train and often-times specific to a single measurement matrix. It was recently demonstrated that iterative sparse-signal-recovery algorithms can be "unrolled" to form interpretable deep networks. Taking inspiration from this work, we develop a novel neural network architecture that mimics the behavior of the denoising-based approximate message passing (D-AMP) algorithm. We call this new network Learned D-AMP (LDAMP). The LDAMP network is easy to train, can be applied to a variety of different measurement matrices, and comes with a state-evolution heuristic that accurately predicts its performance. Most importantly, it outperforms the state-of-the-art BM3D-AMP and NLR-CS algorithms in terms of both accuracy and run time. At high resolutions, and when used with sensing matrices that have fast implementations, LDAMP runs over $50\times$ faster than BM3D-AMP and hundreds of times faster than NLR-CS.

Ali Mousavi, Richard G. Baraniuk

The promise of compressive sensing (CS) has been offset by two significant challenges. First, real-world data is not exactly sparse in a fixed basis. Second, current high-performance recovery algorithms are slow to converge, which limits CS to either non-real-time applications or scenarios where massive back-end computing is available. In this paper, we attack both of these challenges head-on by developing a new signal recovery framework we call {\em DeepInverse} that learns the inverse transformation from measurement vectors to signals using a {\em deep convolutional network}. When trained on a set of representative images, the network learns both a representation for the signals (addressing challenge one) and an inverse map approximating a greedy or convex recovery algorithm (addressing challenge two). Our experiments indicate that the DeepInverse network closely approximates the solution produced by state-of-the-art CS recovery algorithms yet is hundreds of times faster in run time. The tradeoff for the ultrafast run time is a computationally intensive, off-line training procedure typical to deep networks. However, the training needs to be completed only once, which makes the approach attractive for a host of sparse recovery problems.

Ali Mousavi, Arian Maleki, Richard G. Baraniuk

We consider the problem of recovering a vector $β_o \in \mathbb{R}^p$ from $n$ random and noisy linear observations $y= Xβ_o + w$, where $X$ is the measurement matrix and $w$ is noise. The LASSO estimate is given by the solution to the optimization problem $\hatβ_λ = \arg \min_β \frac{1}{2} \|y-Xβ\|_2^2 + λ\| β\|_1$. Among the iterative algorithms that have been proposed for solving this optimization problem, approximate message passing (AMP) has attracted attention for its fast convergence. Despite significant progress in the theoretical analysis of the estimates of LASSO and AMP, little is known about their behavior as a function of the regularization parameter $λ$, or the thereshold parameters $τ^t$. For instance the following basic questions have not yet been studied in the literature: (i) How does the size of the active set $\|\hatβ^λ\|_0/p$ behave as a function of $λ$? (ii) How does the mean square error $\|\hatβ_λ - β_o\|_2^2/p$ behave as a function of $λ$? (iii) How does $\|β^t - β_o \|_2^2/p$ behave as a function of $τ^1, \ldots, τ^{t-1}$? Answering these questions will help in addressing practical challenges regarding the optimal tuning of $λ$ or $τ^1, τ^2, \ldots$. This paper answers these questions in the asymptotic setting and shows how these results can be employed in deriving simple and theoretically optimal approaches for tuning the parameters $τ^1, \ldots, τ^t$ for AMP or $λ$ for LASSO. It also explores the connection between the optimal tuning of the parameters of AMP and the optimal tuning of LASSO.

Ali Mousavi, Arian Maleki, Richard G. Baraniuk

This paper concerns the performance of the LASSO (also knows as basis pursuit denoising) for recovering sparse signals from undersampled, randomized, noisy measurements. We consider the recovery of the signal $x_o \in \mathbb{R}^N$ from $n$ random and noisy linear observations $y= Ax_o + w$, where $A$ is the measurement matrix and $w$ is the noise. The LASSO estimate is given by the solution to the optimization problem $x_o$ with $\hat{x}_λ = \arg \min_x \frac{1}{2} \|y-Ax\|_2^2 + λ\|x\|_1$. Despite major progress in the theoretical analysis of the LASSO solution, little is known about its behavior as a function of the regularization parameter $λ$. In this paper we study two questions in the asymptotic setting (i.e., where $N \rightarrow \infty$, $n \rightarrow \infty$ while the ratio $n/N$ converges to a fixed number in $(0,1)$): (i) How does the size of the active set $\|\hat{x}_λ\|_0/N$ behave as a function of $λ$, and (ii) How does the mean square error $\|\hat{x}_λ - x_o\|_2^2/N$ behave as a function of $λ$? We then employ these results in a new, reliable algorithm for solving LASSO based on approximate message passing (AMP).