Searcharxiv⌕ Search

arXiv subjects

Lovedeep Gondara

Publications and source records attributed to Lovedeep Gondara.

At least 19 recordsLinked to original sources

Right Token, Right Time: From Per-Token Gating to Trajectory-Aware Activation Steering

Activation steering via sparse autoencoders (SAEs) enables behavioral control of large language models without task-specific fine-tuning. Yet, standard methods apply the steering signal at every generated token, displacing representations from the distribution the model was trained on and eroding generation quality. We show that \emph{when} along a generation the intervention is applied is a decisive axis of control. We cast steering as a per-token allocation problem under a fixed budget of steerable tokens, so that methods differ only in which tokens they select, making allocation the sole variable. Within this framework we introduce two gates. The first, Contrastive Feature Projection (CFP), is memoryless and selects the tokens least aligned with the target behavior through a sparse, interpretable feature projection. The second, Closed-Loop Adaptive Bernoulli Steering (CLABS), is trajectory aware. It treats allocation as closed-loop feedback control, tracking the alignment a generation has accumulated so far and spending budget where the trajectory falls behind. On a linear steering-response model we derive an explicit gain region that guarantees closed-loop stability of CLABS and show that its integral term reduces the steady-state tracking offset. Empirically, CFP and CLABS achieve the largest behavioral shift among budgeted gates at matched budgets, with CLABS strongest under the tight budgets where allocation matters most, while best preserving fluency.

cs.LG↗

Reject, Remember, Retry: Why Multi-Turn Attackers Should Remember Their Mistakes

Existing multi-turn jailbreak attackers condition on the same conversation history as the target. We introduce $R^3$ (Reject, Remember, Retry), a black-box multi-LLM framework in which the attacker conditions on a compact state rather than the raw transcript. Turns that fail to raise the attack risk score are removed from the target's context but remembered by the attacker, together with the reason for failure. We call this asymmetric memory. Across two benchmarks and six target LLMs, $R^3$ outperforms state-of-the-art baselines by up to 53% in ASR@5 under matched query budgets, and a low-capacity attacker with $R^3$'s state is able to outperform high-capacity attackers without it by up to 34% in ASR@5. Ablations attribute most gains to remembering rejected turns, and the risk score acts as a coarse threshold rather than a progress measure.

cs.AI↗

Supervised Learning Objectives for High-Dimensional Treatment Effects

Predicting the effect of interventions with many possible variations, e.g., therapeutic content that affects mental health outcomes or an earnings call transcript that drives movement in share price, is useful across several domains. However, classical causal estimators tend to assume that all possible interventions are observed, which is infeasible when interventions vary widely, for instance, in the space of all text strings. We adapt a well-known approach of recasting causal inference as a learning problem to address such settings, assuming realizability of the learned function and no unobserved confounding. We show that causal error decomposes into moment-balancing errors of increasing order and exploit structure of the conditional outcome in the confounders to design objectives that directly improve causal estimation. We also show how to project the effect of a high-dimensional treatment onto coarser attributes, which allows a single model to answer several causal questions without additional attribute-specific training. We empirically evaluate our estimators in settings with high-dimensional continuous, discrete, and text treatments. Our experiments demonstrate generalization to held-out treatments, benefits of balancing higher-order moments when required, and projected causal estimates that are competitive with attribute-specific ones.

cs.LG↗

When Judges Agree and Are Wrong: Consistency-Accuracy Dissociation as a Blind Spot for Label-Free Evaluation

Label-free evaluation methods, such as self-consistency filtering, majority-vote aggregation, and agreement-based reliability scoring, rely on an implicit assumption that consistency can serve as a proxy for correctness when ground-truth labels are unavailable or expensive to obtain. In this paper, we test this assumption directly. Using four financial QA datasets (multiple-choice and free-form), ten consistency dimensions, and two frontier model families, we measure the joint distribution of consistency and accuracy at the level of individual examples rather than in aggregate. We find that 60% of all incorrect answers fall in the high-consistency region (consistency $\geq$ 0.8), and that 77.7% of these show complete answer collapse, i.e. identical wrong answers across all perturbation variants. This high-consistency/incorrect quadrant cannot be detected by any evaluation signal that is built on agreement alone, as the model appears maximally reliable while being wrong. We show that the phenomenon is robust to the choice of consistency threshold t (it persists from t=0.50 to t=1.00), is shared across model families, and is concentrated in specific intersections of perturbation dimension and domain. We map where this quadrant occurs, analyze its stability properties, and discuss the implications for evaluation pipelines that use agreement-based signals in place of human labels.

cs.CL↗

The Format is the Filter: Scoring Format Fixes What an LLM Judge Can Express

When an LLM judge filters instruction-tuning data, its scoring format (binary verdict, 1--5 rating, 0--100 score, or comparison) is rarely justified or ablated. Holding the candidate pool, judge prompt, decoding and selection budget fixed, we score 15,406 UltraFeedback candidates under five formats with three judges, curate, fine-tune 66 models, and evaluate them with four judges. A 0--100 scale is worth only 7--12 equally used levels, two to three times a 1--5 scale rather than twenty, and a binary verdict leaves the top score tied in 77--78\% of prompts, so most selections are made at random. These properties are set by the format almost independently of the judge. Kendall's $τ_b$ hides them: it ranks binary best while $τ_a$ and top-1 agreement rank it last, because it discounts the ties a selector must break at random. Both comparative formats inherit a slot bias whose sign differs across judges. Downstream, curated data beat random selection under every evaluation judge, robust to length and markdown adjustment, but no ordering among formats is detectable. We propose four quantities that judge-based curation pipelines should report.

cs.AI↗

A Coin Flip Per Token: Bernoulli Sparse Steering of Large Language Models

Activation steering via sparse autoencoders (SAEs) enables behavioral control of large language models without task-specific fine-tuning, but standard methods apply the steering signal at every generated token, incurring constant per-token perturbation that risks degrading fluency. We ask: is dense intervention necessary? We introduce Stochastic Token Steering (STS), which gates each token independently with probability $p$, and Stochastic Block Steering (SBS), which gates a leading window once per sequence; neither requires a reward model or learned gating policy. Across two model families and two behavioral tasks, steering only a fraction of tokens recovers most of the behavioral effect of dense steering while incurring less fluency degradation. Across our experimental settings, we further find that the best-performing steering magnitude consistently rises as the intervention ratio falls.

cs.LG↗

Learning from Lost Provenance: Multiple Instance Learning for Cancer Registry Tumor Group Classification

Modernizing cancer registries with deep learning is opening new opportunities to automate labor-intensive tasks such as the coding of pathology reports. However, progress is constrained by the scarcity of report-level human-annotated training data. Cancer registries generate substantial volumes of expert-assigned labels as a routine product of their operations, but these exist at the patient level and are not linked to the individual pathology reports that informed them, limiting their direct use for training models. We develop an efficient framework for training deep learning classifiers by leveraging these operationally-generated labels without requiring per-report human annotation, demonstrated for tumor group classification at the BC Cancer Registry. We use Attention-Based Multiple Instance Learning (ABMIL) to recover the lost link between patient-level labels and the reports that informed them, leveraging the attention the model places on each report to distil a large, noisily-labeled corpus into a compact, high-quality per-report training dataset. A classifier fine-tuned on a distilled dataset achieved a macro F1 of 0.83, outperforming established baselines across most tumor groups. By turning routine operational labels into high-quality training data without additional annotation or large-scale computing infrastructure, ABMIL offers a practical and accessible route to automating cancer registry workflows.

cs.CL↗

De Jure: Iterative LLM Self-Refinement for Structured Extraction of Regulatory Rules

Regulatory documents encode legally binding obligations that LLM-based systems must respect. Yet converting dense, hierarchically structured legal text into machine-readable rules remains a costly, expert-intensive process. We present De Jure, a fully automated, domain-agnostic pipeline for extracting structured regulatory rules from raw documents, requiring no human annotation, domain-specific prompting, or annotated gold data. De Jure operates through four sequential stages: normalization of source documents into structured Markdown; LLM-driven semantic decomposition into structured rule units; multi-criteria LLM-as-a-judge evaluation across 19 dimensions spanning metadata, definitions, and rule semantics; and iterative repair of low-scoring extractions within a bounded regeneration budget, where upstream components are repaired before rule units are evaluated. We evaluate De Jure across four models on three regulatory corpora spanning finance, healthcare, and AI governance. On the finance domain, De Jure yields consistent and monotonic improvement in extraction quality, reaching peak performance within three judge-guided iterations. De Jure generalizes effectively to healthcare and AI governance, maintaining high performance across both open- and closed-source models. In a downstream compliance question-answering evaluation via RAG, responses grounded in De Jure extracted rules are preferred over prior work in 73.8% of cases at single-rule retrieval depth, rising to 84.0% under broader retrieval, confirming that extraction fidelity translates directly into downstream utility. These results demonstrate that explicit, interpretable evaluation criteria can substitute for human annotation in complex regulatory domains, offering a scalable and auditable path toward regulation-grounded LLM alignment.

cs.AI↗

ELM: A Hybrid Ensemble of Language Models for Automated Tumor Group Classification in Population-Based Cancer Registries

Background: Population-based cancer registries (PBCRs) manually extract data from unstructured pathology reports, a labor-intensive process where assigning reports to tumor groups can consume 900 person-hours annually for approximately 100,000 reports at a medium-sized registry. Current automated rule-based systems fail to handle the linguistic complexity of this classification task. Materials and Methods: We present ELM (Ensemble of Language Models), a novel hybrid approach combining small, encoder only language models and large language models (LLMs). ELM employs an ensemble of six fine-tuned encoder only models: three analyzing the top portion and three analyzing the bottom portion of each report to maximize text coverage given token limits. A tumor group is assigned when at least five of six models agree; otherwise, an LLM arbitrates using a carefully curated prompt constrained to likely tumor groups. Results: On a held-out test set of 2,058 pathology reports spanning 19 tumor groups, ELM achieves weighted precision and recall of 0.94, representing a statistically significant improvement (p<0.001) over encoder-only ensembles (0.91 F1-score) and substantially outperforming rule-based approaches. ELM demonstrates particular gains for challenging categories including leukemia (F1: 0.76 to 0.88), lymphoma (0.76 to 0.89), and skin cancer (0.44 to 0.58). Discussion: Deployed in production at British Columbia Cancer Registry, ELM has reduced manual review requirements by approximately 60-70%, saving an estimated 900 person-hours annually while maintaining data quality standards. Conclusion: ELM represents the first successful deployment of a hybrid small, encoder only models-LLM architecture for tumor group classification in a real-world PBCR setting, demonstrating how strategic combination of language models can achieve both high accuracy and operational efficiency.

cs.CL↗

Adapting Natural Language Processing Models Across Jurisdictions: A pilot Study in Canadian Cancer Registries

Population-based cancer registries depend on pathology reports as their primary diagnostic source, yet manual abstraction is resource-intensive and contributes to delays in cancer data. While transformer-based NLP systems have improved registry workflows, their ability to generalize across jurisdictions with differing reporting conventions remains poorly understood. We present the first cross-provincial evaluation of adapting BCCRTron, a domain-adapted transformer model developed at the British Columbia Cancer Registry, alongside GatorTron, a biomedical transformer model, for cancer surveillance in Canada. Our training dataset consisted of approximately 104,000 and 22,000 de-identified pathology reports from the Newfoundland & Labrador Cancer Registry (NLCR) for Tier 1 (cancer vs. non-cancer) and Tier 2 (reportable vs. non-reportable) tasks, respectively. Both models were fine-tuned using complementary synoptic and diagnosis focused report section input pipelines. Across NLCR test sets, the adapted models maintained high performance, demonstrating transformers pretrained in one jurisdiction can be localized to another with modest fine-tuning. To improve sensitivity, we combined the two models using a conservative OR-ensemble achieving a Tier 1 recall of 0.99 and reduced missed cancers to 24, compared with 48 and 54 for the standalone models. For Tier 2, the ensemble achieved 0.99 recall and reduced missed reportable cancers to 33, compared with 54 and 46 for the individual models. These findings demonstrate that an ensemble combining complementary text representations substantially reduce missed cancers and improve error coverage in cancer-registry NLP. We implement a privacy-preserving workflow in which only model weights are shared between provinces, supporting interoperable NLP infrastructure and a future pan-Canadian foundation model for cancer pathology and registry workflows.

cs.CL↗

Beyond Ethics: How Inclusive Innovation Drives Economic Returns in Medical AI

While ethical arguments for fairness in healthcare AI are well-established, the economic and strategic value of inclusive design remains underexplored. This perspective introduces the ``inclusive innovation dividend'' -- the counterintuitive principle that solutions engineered for diverse, constrained use cases generate superior economic returns in broader markets. Drawing from assistive technologies that evolved into billion-dollar mainstream industries, we demonstrate how inclusive healthcare AI development creates business value beyond compliance requirements. We identify four mechanisms through which inclusive innovation drives returns: (1) market expansion via geographic scalability and trust acceleration; (2) risk mitigation through reduced remediation costs and litigation exposure; (3) performance dividends from superior generalization and reduced technical debt, and (4) competitive advantages in talent acquisition and clinical adoption. We present the Healthcare AI Inclusive Innovation Framework (HAIIF), a practical scoring system that enables organizations to evaluate AI investments based on their potential to capture these benefits. HAIIF provides structured guidance for resource allocation, transforming fairness and inclusivity from regulatory checkboxes into sources of strategic differentiation. Our findings suggest that organizations investing incrementally in inclusive design can achieve expanded market reach and sustained competitive advantages, while those treating these considerations as overhead face compounding disadvantages as network effects and data advantages accrue to early movers.

cs.AI↗

Small or Large? Zero-Shot or Finetuned? Guiding Language Model Choice for Specialized Applications in Healthcare

This study aims to guide language model selection by investigating: 1) the necessity of finetuning versus zero-shot usage, 2) the benefits of domain-adjacent versus generic pretrained models, 3) the value of further domain-specific pretraining, and 4) the continued relevance of Small Language Models (SLMs) compared to Large Language Models (LLMs) for specific tasks. Using electronic pathology reports from the British Columbia Cancer Registry (BCCR), three classification scenarios with varying difficulty and data size are evaluated. Models include various SLMs and an LLM. SLMs are evaluated both zero-shot and finetuned; the LLM is evaluated zero-shot only. Finetuning significantly improved SLM performance across all scenarios compared to their zero-shot results. The zero-shot LLM outperformed zero-shot SLMs but was consistently outperformed by finetuned SLMs. Domain-adjacent SLMs generally performed better than the generic SLM after finetuning, especially on harder tasks. Further domain-specific pretraining yielded modest gains on easier tasks but significant improvements on the complex, data-scarce task. The results highlight the critical role of finetuning for SLMs in specialized domains, enabling them to surpass zero-shot LLM performance on targeted classification tasks. Pretraining on domain-adjacent or domain-specific data provides further advantages, particularly for complex problems or limited finetuning data. While LLMs offer strong zero-shot capabilities, their performance on these specific tasks did not match that of appropriately finetuned SLMs. In the era of LLMs, SLMs remain relevant and effective, offering a potentially superior performance-resource trade-off compared to LLMs.

cs.CL↗

Bridging AI Innovation and Healthcare Needs: Lessons Learned from Incorporating Modern NLP at The BC Cancer Registry

Automating data extraction from clinical documents offers significant potential to improve efficiency in healthcare settings, yet deploying Natural Language Processing (NLP) solutions presents practical challenges. Drawing upon our experience implementing various NLP models for information extraction and classification tasks at the British Columbia Cancer Registry (BCCR), this paper shares key lessons learned throughout the project lifecycle. We emphasize the critical importance of defining problems based on clear business objectives rather than solely technical accuracy, adopting an iterative approach to development, and fostering deep interdisciplinary collaboration and co-design involving domain experts, end-users, and ML specialists from inception. Further insights highlight the need for pragmatic model selection (including hybrid approaches and simpler methods where appropriate), rigorous attention to data quality (representativeness, drift, annotation), robust error mitigation strategies involving human-in-the-loop validation and ongoing audits, and building organizational AI literacy. These practical considerations, generalizable beyond cancer registries, provide guidance for healthcare organizations seeking to successfully implement AI/NLP solutions to enhance data management processes and ultimately improve patient care and public health outcomes.

cs.CL↗

Leveraging Language Models for Automated Patient Record Linkage

Objective: Healthcare data fragmentation presents a major challenge for linking patient data, necessitating robust record linkage to integrate patient records from diverse sources. This study investigates the feasibility of leveraging language models for automated patient record linkage, focusing on two key tasks: blocking and matching. Materials and Methods: We utilized real-world healthcare data from the Missouri Cancer Registry and Research Center, linking patient records from two independent sources using probabilistic linkage as a baseline. A transformer-based model, RoBERTa, was fine-tuned for blocking using sentence embeddings. For matching, several language models were experimented under fine-tuned and zero-shot settings, assessing their performance against ground truth labels. Results: The fine-tuned blocking model achieved a 92% reduction in the number of candidate pairs while maintaining near-perfect recall. In the matching task, fine-tuned Mistral-7B achieved the best performance with only 6 incorrect predictions. Among zero-shot models, Mistral-Small-24B performed best, with a total of 55 incorrect predictions. Discussion: Fine-tuned language models achieved strong performance in patient record blocking and matching with minimal errors. However, they remain less accurate and efficient than a hybrid rule-based and probabilistic approach for blocking. Additionally, reasoning models like DeepSeek-R1 are impractical for large-scale record linkage due to high computational costs. Conclusion: This study highlights the potential of language models for automating patient record linkage, offering improved efficiency by eliminating the manual efforts required to perform patient record linkage. Overall, language models offer a scalable solution that can enhance data integration, reduce manual effort, and support disease surveillance and research.

cs.AI↗

A Clinical Trial Design Approach to Auditing Language Models in Healthcare Setting

We present an audit mechanism for language models, with a focus on models deployed in the healthcare setting. Our proposed mechanism takes inspiration from clinical trial design where we posit the language model audit as a single blind equivalence trial, with the comparison of interest being the subject matter experts. We show that using our proposed method, we can follow principled sample size and power calculations, leading to the requirement of sampling minimum number of records while maintaining the audit integrity and statistical soundness. Finally, we provide a real-world example of the audit used in a production environment in a large-scale public health network.

cs.CY↗

Differentially Private Ensemble Classifiers for Data Streams

Learning from continuous data streams via classification/regression is prevalent in many domains. Adapting to evolving data characteristics (concept drift) while protecting data owners' private information is an open challenge. We present a differentially private ensemble solution to this problem with two distinguishing features: it allows an \textit{unbounded} number of ensemble updates to deal with the potentially never-ending data streams under a fixed privacy budget, and it is \textit{model agnostic}, in that it treats any pre-trained differentially private classification/regression model as a black-box. Our method outperforms competitors on real-world and simulated datasets for varying settings of privacy, concept drift, and data distribution.

cs.LG↗

The Differentially Private Lottery Ticket Mechanism

We propose the differentially private lottery ticket mechanism (DPLTM). An end-to-end differentially private training paradigm based on the lottery ticket hypothesis. Using "high-quality winners", selected via our custom score function, DPLTM significantly improves the privacy-utility trade-off over the state-of-the-art. We show that DPLTM converges faster, allowing for early stopping with reduced privacy budget consumption. We further show that the tickets from DPLTM are transferable across datasets, domains, and architectures. Our extensive evaluation on several public datasets provides evidence to our claims.

cs.LG↗

Differentially Private Survival Function Estimation

Survival function estimation is used in many disciplines, but it is most common in medical analytics in the form of the Kaplan-Meier estimator. Sensitive data (patient records) is used in the estimation without any explicit control on the information leakage, which is a significant privacy concern. We propose a first differentially private estimator of the survival function and show that it can be easily extended to provide differentially private confidence intervals and test statistics without spending any extra privacy budget. We further provide extensions for differentially private estimation of the competing risk cumulative incidence function, Nelson-Aalen's estimator for the hazard function, etc. Using eleven real-life clinical datasets, we provide empirical evidence that our proposed method provides good utility while simultaneously providing strong privacy guarantees.

cs.LG↗