Clemson University

4 related articles

Subscribe to updates

arXiv Nov 5, 2026
ProtoGestalt ties rare-disease face screening to eyebrows, eyes, nose and cheeks, reaching 0.85 accuracy on matched controls
Rare genetic disorders are often missed at first presentation, and most facial diagnostic models are end-to-end black boxes that cannot say what drove a call. ProtoGestalt adds facial-landmark constraints to PIP-Net prototype learning, so each prototype draws evidence only from one of ten anatomical regions and predictions are sums of those local prototypes. On controls matched by age, sex and ethnicity it separated syndromic from unaffected faces at 0.85 accuracy, versus 0.73 for the region-masked variant; AUPRC fell as controls increased but stayed above 0.70.
Rare genetic disorders are often missed at first presentation, and most facial diagnostic models are end-to-end black boxes that cannot say what drove a call. ProtoGestalt adds facial-landmark constraints to PIP-Net prototype learning, so each prototype draws evidence only from one of ten anatomical regions and predictions are sums of those local prototypes. On controls matched by age, sex and ethnicity it separated syndromic from unaffected faces at 0.85 accuracy, versus 0.73 for the region-masked variant; AUPRC fell as controls increased but stayed above 0.70.
Clemson University University of Augsburg University of Bonn
Ranga Baminiwatte（First author）
ProtoGestalt Prototype Learning Facial Phenotyping
arXiv Oct 14, 2026
Self-supervised learning stops at shared content: record each instance's augmentation strength and the coordinates line up
Recording one augmentation strength per instance and letting content variances respond to it lets self-supervised learning separate content coordinates that previously stayed mixed, raising dSprites factor alignment from 0.837 to 0.932.
Clemson University
D. Hudson Smith（First author, Corresponding author）
Variance Fingerprints Component Identification in Self-Supervised Learning Self-Supervised Learning dSprites
arXiv Oct 2, 2026
LinkedIn and university collaborators find that 7B-to-72B students trained on rejects from smaller frozen models do better on code and math while using less compute
Holding prompts, chosen responses, SeqKD initialization, DPO objective, and training budget fixed while varying only the reject source, the study finds that across Qwen2.5 students from 7B to 72B, rejects generated by smaller frozen Base models with less inference compute train stronger students than the students' own rejects on code generation and mathematical reasoning, and it derives a finite-horizon utility bound in a linearized feature model that motivates three interventions: source mixing, prompt reassignment with lexical permutation, and reference-likelihood reselection.
Holding prompts, chosen responses, SeqKD initialization, DPO objective, and training budget fixed while varying only the reject source, the study finds that across Qwen2.5 students from 7B to 72B, rejects generated by smaller frozen Base models with less inference compute train stronger students than the students' own rejects on code generation and mathematical reasoning, and it derives a finite-horizon utility bound in a linearized feature model that motivates three interventions: source mixing, prompt reassignment with lexical permutation, and reference-likelihood reselection.
University of Wisconsin–Madison Clemson University Arizona State University Pennsylvania State University University of California, Davis
Rui Cai（First author）
Preference Distillation Reject Source Selection Direct Preference Optimization Code Generation Mathematical Reasoning
bioRxiv Sep 17, 2026
The ModelSEED Biochemistry Database, 2026 update: grading multi-source thermodynamics
This update expands the ModelSEED Biochemistry Database to roughly 46,000 compounds, 56,000 reactions and 37,000 metabolic structures, widens thermodynamic handling from two sources to four (group contribution, eQuilibrator 3.0, dGPredictor and experimental values) each kept with its own uncertainty, assigns gold, silver or bronze evidence grades to about 33,000 reactions, and releases reaction directions predicted by an ensemble of large language models alongside a new conflict-resolution pipeline that documents structural choices across sources.
This update expands the ModelSEED Biochemistry Database to roughly 46,000 compounds, 56,000 reactions and 37,000 metabolic structures, widens thermodynamic handling from two sources to four (group contribution, eQuilibrator 3.0, dGPredictor and experimental values) each kept with its own uncertainty, assigns gold, silver or bronze evidence grades to about 33,000 reactions, and releases reaction directions predicted by an ensemble of large language models alongside a new conflict-resolution pipeline that documents structural choices across sources.
Northwestern University University of Chicago Weizmann Institute of Science Clemson University University of Potsdam Technische Universität Berlin University of Illinois Urbana-Champaign Pennsylvania State University
Andrew Freiburger（First author）Samuel M. D. Seaver（Corresponding author）
ModelSEED Biochemistry Database Multi-source thermodynamics evidence grading Atom mapping Large language models

You’re all caught up

ProtoGestalt ties rare-disease face screening to eyebrows, eyes, nose and cheeks, reaching 0.85 accuracy on matched controls

Self-supervised learning stops at shared content: record each instance's augmentation strength and the coordinates line up

LinkedIn and university collaborators find that 7B-to-72B students trained on rejects from smaller frozen models do better on code and math while using less compute

The ModelSEED Biochemistry Database, 2026 update: grading multi-source thermodynamics