Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Shariq Bashir has a total of 14 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    A survey of theory-grounded interpretability for deep neural networks

    by Muhammad Usman Khalid, Abeer A. K. Alharbi, Shariq Bashir, Shafiq Ur Rehman

    Deep neural networks achieve state-of-the-art performance across computer vision, natural language processing, and scientific discovery, yet their internal decision-making mechanisms remain difficult to interpret. Post-hoc explanation methods such as SHAP, saliency maps, and attention visualization provide empirical insights but offer limited mathematical guarantees about learning dynamics and representation formation. This survey takes a complementary, theory-grounded perspective, examining interpretability through four classical machine learning (ML) frameworks, each grounded in a mature mathematical theory. These are kernel methods (functional analysis), sparse representations (convex optimization), matrix factorization (linear algebra), and manifold learning (differential geometry). We synthesize and critically evaluate 203 papers (2016–2026), organizing them by the theoretical principle each framework contributes rather than by application domain. Neural tangent kernel (NTK) methods provide rigorous convergence and generalization guarantees but depend on idealized infinite-width assumptions; sparse representation methods yield interpretable-by-design architectures but face reconstruction and approximation trade-offs; matrix factorization reveals an implicit bias toward low-rank structure that helps explain generalization; and manifold learning offers geometric tools for analyzing representation structure, class separability, and generative behavior. Beyond reviewing each framework, we establish explicit mathematical bridges between them, arguing that they offer complementary views of a common phenomenon, namely that gradient-based optimization often drives deep networks toward low-dimensional, interpretable structure. We further situate these frameworks among adjacent interpretability paradigms, including mechanistic, concept-based, and attribution-based methods, and delineate the validity conditions and failure modes that determine when classical theory yields reliable rather than misleading explanations. The review concludes with open problems and directions for scalable, theory-grounded interpretability.

  • Open Access

    XAI-DR-GAN: A topology-aware explainable GAN framework for interpretable diabetic retinopathy grading

    by Pulkit Dwivedi, Soumendu Chakraborty, Niyaz Ahmad Wani, Mudasir Ahmad Wani, Shariq Bashir, Kashish Ara Shakil


  • Open Access

    User-Controlled Post-Hoc Watermarking for LLM-Generated Text Without Model Access

    by Shariq Bashir, Akmal Saeed Khattak

    Current watermarking approaches for LLM-generated text typically embed watermarks during the generation process, using a secret key controlled by the LLM. Although this enables verification of LLM-generated content, it may raise copyright and authorship concerns. For instance, when a user relies on an LLM for proofreading, grammar correction, or translation, the model may still embed its own watermark. This can later create disputes regarding ownership, as the text may appear to belong to the AI system rather than the human author. To address this issue, this paper proposes an improved post-hoc watermarking approach that embeds user-controlled watermarks after the text generation is complete. The approach does not require access to the internal architecture of the LLM. It applies a probabilistic watermarking mechanism using a user secret key, and embeds watermark signals across the entire text rather than only on selective terms, as in previous post-hoc approaches. During watermark embedding, the approach regenerates the suffix after each token substitution by interacting with the LLM. This preserves grammatical correctness and contextual coherence, addressing the quality degradation often observed in synonym-based substitution approaches that do not adjust the surrounding text. A BERT-based masked language model is used to ensure that each substitution remains semantically consistent with the original text. Experimental results show that the proposed approach maintains higher similarity and readability than previous post-hoc approaches while producing statistically verifiable watermark signals. The approach provides a practical solution for authorship verification and copyright protection in generative AI, without requiring access to the internal mechanism of the LLM.


  • Open Access

    A pattern-growth approach for mining maximal fault-tolerant frequent itemsets

    by Shariq Bashir

    Mining fault-tolerant (FT) frequent itemsets in noisy datasets is more challenging than conventional frequent itemset mining due to the high cost of evaluating fault-tolerance conditions. Consequently, mining maximal fault-tolerant frequent itemsets (FT-MFIs) is particularly important, as they provide a concise representation of all frequent patterns while eliminating redundancy, which is crucial in noisy datasets where error tolerance must be considered. Existing approaches for mining FT-MFIs are predominantly based on Apriori-style candidate generation and test, which suffer from exponential candidate growth, multiple dataset scans, and limited scalability. As a novel contribution, this work is the first to propose a pattern-growth framework for mining FT-MFIs. We introduce an algorithm that constructs a fault-tolerant FP-tree (FT-FP-tree) to compress transactions with common prefixes and evaluate tolerance conditions in a single pass. Additionally, new techniques for transaction mapping and conditional pattern extraction enhance the efficiency and scalability of the mining process. Experimental results on benchmark datasets demonstrate that the proposed approach achieves substantial reductions in execution time compared to all existing algorithms, while also showing improved memory efficiency relative to other pattern-growth based algorithms. This establishes pattern growth as a practical and scalable solution for fault-tolerant frequent itemset discovery.


  • Multi-features Point-of-Interest (POI) recommendation using graph neural network

    by Anwar Sadad, Akmal Saeed Khattak, Rabeeh Ayaz Abbasi, Muddassar Azam Sindhu, Shariq Bashir


  • Open Access

    Zero-Shot Pre-Retrieval Prompt Performance Prediction for Large Language Models

    by Shariq Bashir

    Query Performance Prediction (QPP) plays a central role in information retrieval by estimating how effectively a query will perform before retrieval. However, traditional pre-retrieval QPPs rely on corpus-level statistics such as term frequency or document frequency information that is unavailable in the context of Large Language Models (LLMs), where model parameters and training data are inaccessible. In addition, post-response analysis across many LLMs is computationally costly and impractical at scale. To overcome these challenges, this work introduces a novel zero-shot framework for Prompt Performance Prediction, designed to estimate an LLM’s expertise in handling a given prompt without relying on any generated responses. The framework proposes a new family of prompt predictors that capture the lexical, topical, intentional, and cognitive dimensions of the prompt text. These predictors, collectively referred to as Prompt Performance Predictors, enable pre-response estimation of model suitability based solely on the textual characteristics of the prompt. In addition, the study explores the integration of multiple predictors within a hybrid neural learning framework to assess their individual and combined contributions. Experimental evaluation on 1,130 LLMs from the TREC Million LLM Track demonstrates that the proposed prompt predictors significantly outperform traditional pre-retrieval QPPs in predicting model expertise. The results further reveal that combining diverse predictors yields the highest overall effectiveness, confirming that prompt-level linguistic, semantic, and intent features provide complementary and reliable signals for LLM expertise prediction.

  • Open Access

    Asymmetric Watermarking for Large Language Models With Public and Private Verification

    by Shariq Bashir

    The rapid proliferation of large language models (LLMs) has intensified the need for reliable mechanisms to identify, authenticate, and trace AI-generated text. LLM watermarking addresses this challenge by embedding imperceptible statistical signals into generated content, enabling provenance verification and misuse detection. However, existing watermarking approaches rely on a symmetric secret key shared between watermark embedding and detection, which prevents public verification and introduces a critical single point of failure: key disclosure enables watermark forgery, spoofing, and unauthorized replication. This paper proposes an asymmetric watermarking approach for LLMs that decouples watermark generation from verification using a public–private key pair. In the proposed design, a private key is used exclusively by the LLM to embed the watermark, while a corresponding public key enables anyone to verify the watermark without compromising security. The approach extends probabilistic token biasing by jointly embedding private and public watermark signals, allowing transparent public verification while preserving strong private authentication guarantees. Experimental evaluation demonstrates that the proposed approach closely matches the statistical behavior of the traditional symmetric watermark under the private key, while introducing a robust and independently verifiable public watermark. High correlations in z-scores and p-values confirm equivalent detectability, and BERT-based similarity scores show that semantic quality is preserved across a wide range of watermark strengths. These results demonstrate that the proposed asymmetric approach achieves statistically significant detection, strong robustness against forgery, and practical public verifiability, making it well suited for forensic and real-world deployment of LLM-generated content.


  • Open Access

    Private Information Retrieval From AI Chatbots via Proxy-Prompt Masking

    by Shariq Bashir

    This paper proposes a privacy-preserving scheme, PromptMasking, for querying AI chatbots without disclosing users’ sensitive prompts. This scheme enables users to query AI systems without revealing their true intent by transforming sensitive prompts into non-sensitive proxy prompts, selected from a semantically constructed proxy dictionary. The scheme then supplements these with cover prompts, ensuring the original prompt remains indistinguishable during processing. Unlike prior techniques, PromptMasking does not rely on encryption or prompt rewriting, making it compatible with existing large language model APIs. To evaluate its effectiveness, we constructed a dataset comprising 240 topics (40 sensitive and 200 non-sensitive) and generated over 1.2 million high-quality natural language prompts. Experimental results demonstrate that PromptMasking significantly outperforms state-of-the-art baselines in preserving prompt privacy. In particular, it achieves substantial improvements in obfuscating related prompts across both single and multi-session search settings.