Erfan Miahi

Reza Khanmohammadi, Erfan Miahi, Simerjot Kaur et al.

The miscalibration of Large Reasoning Models (LRMs) undermines their reliability in high-stakes domains, necessitating methods to accurately estimate the confidence of their long-form, multi-step outputs. To address this gap, we introduce the Reasoning Model Confidence estimation Benchmark (RMCB), a public resource of 347,496 reasoning traces from six popular LRMs across different architectural families. The benchmark is constructed from a diverse suite of datasets spanning high-stakes domains, including clinical, financial, legal, and mathematical reasoning, alongside complex general reasoning benchmarks, with correctness annotations provided for all samples. Using RMCB, we conduct a large-scale empirical evaluation of over ten distinct representation-based methods, spanning sequential, graph-based, and text-based architectures. Our central finding is a persistent trade-off between discrimination (AUROC) and calibration (ECE): text-based encoders achieve the best AUROC (0.672), while structurally-aware models yield the best ECE (0.148), with no single method dominating both. Furthermore, we find that increased architectural complexity does not reliably outperform simpler sequential baselines, suggesting a performance ceiling for methods relying solely on chunk-level hidden states. This work provides the most comprehensive benchmark for this task to date, establishing rigorous baselines and demonstrating the limitations of current representation-based paradigms.

Reza Khanmohammadi, Erfan Miahi, Simerjot Kaur et al.

Large vision-language models suffer from visual ungroundedness: they can produce a fluent, confident, and even correct response driven entirely by language priors, with the image contributing nothing to the prediction. Existing confidence estimation methods cannot detect this, as they observe model behavior under normal inference with no mechanism to determine whether a prediction was shaped by the image or by text alone. We introduce BICR (Blind-Image Contrastive Ranking), a model-agnostic confidence estimation framework that makes this contrast explicit during training by extracting hidden states from a frozen LVLM twice: once with the real image-question pair, and once with the image blacked out while the question is held fixed. A lightweight probe is trained on the real-image hidden state and regularized by a ranking loss that penalizes higher confidence on the blacked-out view, teaching it to treat visual grounding as a signal of reliability at zero additional inference cost. Evaluated across five modern LVLMs and seven baselines on a benchmark covering visual question answering, object hallucination detection, medical imaging, and financial document understanding, BICR achieves the best cross-LVLM average on both calibration and discrimination simultaneously, with statistically significant discrimination gains robust to cluster-aware analysis at 4-18x fewer parameters than the strongest probing baseline.

Muhammad Kamran Janjua, Haseeb Shah, Martha White et al.

In this paper we investigate the use of reinforcement-learning based prediction approaches for a real drinking-water treatment plant. Developing such a prediction system is a critical step on the path to optimizing and automating water treatment. Before that, there are many questions to answer about the predictability of the data, suitable neural network architectures, how to overcome partial observability and more. We first describe this dataset, and highlight challenges with seasonality, nonstationarity, partial observability, and heterogeneity across sensors and operation modes of the plant. We then describe General Value Function (GVF) predictions -- discounted cumulative sums of observations -- and highlight why they might be preferable to classical n-step predictions common in time series prediction. We discuss how to use offline data to appropriately pre-train our temporal difference learning (TD) agents that learn these GVF predictions, including how to select hyperparameters for online fine-tuning in deployment. We find that the TD-prediction agent obtains an overall lower normalized mean-squared error than the n-step prediction agent. Finally, we show the importance of learning in deployment, by comparing a TD agent trained purely offline with no online updating to a TD agent that learns online. This final result is one of the first to motivate the importance of adapting predictions in real-time, for non-stationary high-volume systems in the real world.

Joel Lidin, Amir Sarfi, Erfan Miahi et al.

Recently, there has been increased interest in globally distributed training, which has the promise to both reduce training costs and democratize participation in building large-scale foundation models. However, existing models trained in a globally distributed manner are relatively small in scale and have only been trained with whitelisted participants. Therefore, they do not yet realize the full promise of democratized participation. In this report, we describe Covenant-72B, an LLM produced by the largest collaborative globally distributed pre-training run (in terms of both compute and model scale), which simultaneously allowed open, permissionless participation supported by a live blockchain protocol. We utilized a state-of-the-art communication-efficient optimizer, SparseLoCo, supporting dynamic participation with peers joining and leaving freely. Our model, pre-trained on approximately 1.1T tokens, performs competitively with fully centralized models pre-trained on similar or higher compute budgets, demonstrating that fully democratized, non-whitelisted participation is not only feasible, but can be achieved at unprecedented scale for a globally distributed pre-training run.

Reza Khanmohammadi, Erfan Miahi, Mehrsa Mardikoraem et al.

Miscalibration in Large Language Models (LLMs) undermines their reliability, highlighting the need for accurate confidence estimation. We introduce CCPS (Calibrating LLM Confidence by Probing Perturbed Representation Stability), a novel method analyzing internal representational stability in LLMs. CCPS applies targeted adversarial perturbations to final hidden states, extracts features reflecting the model's response to these perturbations, and uses a lightweight classifier to predict answer correctness. CCPS was evaluated on LLMs from 8B to 32B parameters (covering Llama, Qwen, and Mistral architectures) using MMLU and MMLU-Pro benchmarks in both multiple-choice and open-ended formats. Our results show that CCPS significantly outperforms current approaches. Across four LLMs and three MMLU variants, CCPS reduces Expected Calibration Error by approximately 55% and Brier score by 21%, while increasing accuracy by 5 percentage points, Area Under the Precision-Recall Curve by 4 percentage points, and Area Under the Receiver Operating Characteristic Curve by 6 percentage points, all relative to the strongest prior method. CCPS delivers an efficient, broadly applicable, and more accurate solution for estimating LLM confidence, thereby improving their trustworthiness.

Erfan Miahi, Eugene Belilovsky

Reinforcement learning (RL) is a critical component for post-training large language models (LLMs). However, in bandwidth-constrained distributed RL, scalability is often bottlenecked by the synchronization of policy weights from trainers to inference workers, particularly over commodity networks or in decentralized settings. While recent studies suggest that RL updates modify only a small fraction of model parameters, these observations are typically based on coarse checkpoint differences. We present a systematic empirical study of weight-update sparsity at both step-level and multi-step granularities, examining its evolution across training dynamics, off-policy delay, and model scale. We find that update sparsity is consistently high, frequently exceeding 99% across practically relevant settings. Leveraging this structure, we propose PULSE (Patch Updates via Lossless Sparse Encoding), a simple yet highly efficient lossless weight synchronization method that transmits only the indices and values of modified parameters. PULSE is robust to transmission errors and avoids floating-point drift inherent in additive delta schemes. In bandwidth-constrained decentralized environments, our approach achieves over 100x (14 GB to ~108 MB) communication reduction while maintaining bit-identical training dynamics and performance compared to full weight synchronization. By exploiting this structure, PULSE enables decentralized RL training to approach centralized throughput, reducing the bandwidth required for weight synchronization from 20 Gbit/s to 0.2 Gbit/s to maintain high GPU utilization.

Han Wang, Erfan Miahi, Martha White et al.

In this paper we investigate the properties of representations learned by deep reinforcement learning systems. Much of the early work on representations for reinforcement learning focused on designing fixed-basis architectures to achieve properties thought to be desirable, such as orthogonality and sparsity. In contrast, the idea behind deep reinforcement learning methods is that the agent designer should not encode representational properties, but rather that the data stream should determine the properties of the representation -- good representations emerge under appropriate training schemes. In this paper we bring these two perspectives together, empirically investigating the properties of representations that support transfer in reinforcement learning. We introduce and measure six representational properties over more than 25 thousand agent-task settings. We consider Deep Q-learning agents with different auxiliary losses in a pixel-based navigation environment, with source and transfer tasks corresponding to different goal locations. We develop a method to better understand why some representations work better for transfer, through a systematic approach varying task similarity and measuring and correlating representation properties with transfer performance. We demonstrate the generality of the methodology by investigating representations learned by a Rainbow agent that successfully transfer across games modes in Atari 2600.

Mojtaba Shakeri, Erfan Miahi, Abhishek Gupta et al.

In today's digital world, we are faced with an explosion of data and models produced and manipulated by numerous large-scale cloud-based applications. Under such settings, existing transfer evolutionary optimization frameworks grapple with simultaneously satisfying two important quality attributes, namely (1) scalability against a growing number of source tasks and (2) online learning agility against sparsity of relevant sources to the target task of interest. Satisfying these attributes shall facilitate practical deployment of transfer optimization to scenarios with big task-instances, while curbing the threat of negative transfer. While applications of existing algorithms are limited to tens of source tasks, in this paper, we take a quantum leap forward in enabling more than two orders of magnitude scale-up in the number of tasks; i.e., we efficiently handle scenarios beyond 1000 source task-instances. We devise a novel transfer evolutionary optimization framework comprising two co-evolving species for joint evolutions in the space of source knowledge and in the search space of solutions to the target problem. In particular, co-evolution enables the learned knowledge to be orchestrated on the fly, expediting convergence in the target optimization task. We have conducted an extensive series of experiments across a set of practically motivated discrete and continuous optimization examples comprising a large number of source task-instances, of which only a small fraction indicate source-target relatedness. The experimental results show that not only does our proposed framework scale efficiently with a growing number of source tasks but is also effective in capturing relevant knowledge against sparsity of related sources, fulfilling the two salient features of scalability and online learning agility.

Erfan Miahi, Seyed Abolghasem Mirroshandel, Alexis Nasr

Male infertility is a disease which affects approximately 7% of men. Sperm morphology analysis (SMA) is one of the main diagnosis methods for this problem. Manual SMA is an inexact, subjective, non-reproducible, and hard to teach process. As a result, in this paper, we introduce a novel automatic SMA based on a neural architecture search algorithm termed Genetic Neural Architecture Search (GeNAS). For this purpose, we used a collection of images called MHSMA dataset contains 1,540 sperm images which have been collected from 235 patients with infertility problems. GeNAS is a genetic algorithm that acts as a meta-controller which explores the constrained search space of plain convolutional neural network architectures. Every individual of the genetic algorithm is a convolutional neural network trained to predict morphological deformities in different segments of human sperm (head, vacuole, and acrosome), and its fitness is calculated by a novel proposed method named GeNAS-WF especially designed for noisy, low resolution, and imbalanced datasets. Also, a hashing method is used to save each trained neural architecture fitness, so we could reuse them during fitness evaluation and speed up the algorithm. Besides, in terms of running time and computation power, our proposed architecture search method is far more efficient than most of the other existing neural architecture search algorithms. Additionally, other proposed methods have been evaluated on balanced datasets, whereas GeNAS is built specifically for noisy, low quality, and imbalanced datasets which are common in the field of medical imaging. In our experiments, the best neural architecture found by GeNAS has reached an accuracy of 91.66%, 77.33%, and 77.66% in the vacuole, head, and acrosome abnormality detection, respectively. In comparison to other proposed algorithms for MHSMA dataset, GeNAS achieved state-of-the-art results.