Arman Cohan · page 2
Papers
- Diffusion vs. Autoregressive Language Models: A Text Embedding Perspective (2025 · 10 citations)
Embedding models built on decoder-only LLMs have become very strong. However, they inherit causal attention from pre-training: each token's representation ignores everything that follows it. This is a poor fit for embedding tasks, which benefit from seeing the whole text at once. The authors propose DiffEmbed, which… - AbGen: Evaluating Large Language Models in Ablation Study Design and Evaluation for Scientific Research (2025 · 9 citations)
AbGen tests whether large language models can help scientists plan experiments, focusing on ablation study design. - TESS 2: A Large-Scale Generalist Diffusion Language Model (2025 · 9 citations)
Diffusion language models have mostly stayed small and been judged on intrinsic metrics such as perplexity, so it has been unclear whether they can work as general instruction-following assistants. - SciRAG: Adaptive, Citation-Aware, and Outline-Guided Retrieval and Synthesis for Scientific Literature (2026 · 7 citations)
Scientific output keeps growing quickly, and existing retrieval-augmented generation (RAG) tools for literature QA have four main weaknesses. - MetaFaith: Faithful Natural Language Uncertainty Expression in LLMs (2025 · 6 citations)
LLMs often state false claims in confident language, which leads users to over-trust them. This paper studies "faithful calibration": whether the decisiveness of a model's wording matches its intrinsic uncertainty. This differs from standard factual calibration, which ties confidence to accuracy. The authors… - From Scores to Steps: Diagnosing and Improving LLM Performance in Evidence-Based Medical Calculations (2025 · 5 citations)
LLMs are increasingly considered for clinical decision support, yet their skill at medical calculations, such as GFR or cardiovascular risk scores, is poorly measured. - Patient-Similarity Cohort Reasoning in Clinical Text-to-SQL (2026 · 4 citations)
Clinicians often ask for more than one patient's record. They want to compare groups of similar patients under clinical constraints such as diagnosis codes, time windows and lab ranges. Earlier EHR text-to-SQL datasets like MIMICSQL and EHRSQL mostly target single patients or simple statistics. The authors built… - RbtAct: Rebuttal as Supervision for Actionable Review Feedback Generation (2026 · 2 citations)
LLM-drafted peer reviews often sound fluent but stay generic, so authors get little guidance they can actually implement. - Anchor: Branch-Point Data Generation for GUI Agents (2026 · 1 citation)
Training end-to-end GUI agents for desktop operating systems requires a lot of interaction data. Human demonstrations are costly, and automatic pipelines tend to produce either narrow task sets or noisy trajectories that drift from the goal. Anchor addresses this by growing a large training corpus from a small number… - Can AI Be a Good Peer Reviewer? A Survey of Peer Review Process, Evaluation, and the Future (2026 · 1 citation)
This survey asks how far large language models can support or automate scholarly peer review. It treats review as a full pipeline (reviewing, rebuttal, meta-review, decision, and revision) rather than a single text-generation step. The authors build a taxonomy across three areas: - CPTCoder: A Reliable LLM System for Medical Procedure Code Prediction (2026 · 1 citation)
CPTCoder targets automatic assignment of Current Procedural Terminology (CPT) codes to clinical reports. This is a long-tailed, multi-label task in which codes are five-digit identifiers, so a single wrong digit names a different billable procedure. The authors fine-tune Llama-3.1-8B-Instruct with LoRA. They add…
This page in Markdown → https://md.research.elamur.ai/free/en/authors/arman-cohan/page/2/