Towards Reversible Forgetting: Managing Obsolete Knowledge in Continual Enterprise AI Agents

Continual learning has traditionally treated forgetting as a failure, emphasizing preservation of previously acquired knowledge as environments evolve. We argue that this objective is incomplete for enterprise AI agents operating in non-stationary environments, where customers, policies, tools, workflows, regulations, and market conditions change over time. Indiscriminate retention can allow obsolete knowledge to influence decisions, creating negative transfer and operational risk. We therefore propose reversible forgetting: a conceptual framework with three operational memory states: active, dormant, and retired, and a reactivation transition that can restore dormant knowledge when its relevance returns. We instantiate the framework as a Hysteretic Reversible Memory Controller that accumulates relevance evidence, uses asymmetric thresholds to prevent state oscillation, tests reactivation in shadow mode, and gates retirement through policy. The framework reduces the influence of obsolete information without conflating temporary suppression with permanent erasure. Finance illustrates the idea: knowledge useful under one market regime may become harmful under another yet regain relevance when similar conditions recur.

bioMoR: Biology-Guided Mixture-of-Recursions for Effective Genomic Learning

Transformer models for high-dimensional omics analysis process thousands of genes or pathways, although only a subset requires deep computation. Mixture-of-Recursions (MoR) improves efficiency through adaptive token-choice or expert-choice routing. We propose bioMoR, which, to the best of our knowledge, is the first framework to apply MoR to gene-level and pathway-level learning. Our contributions include identifying three locations for integrating structured biological knowledge within an MoR backbone: graph-based information sharing refines token embeddings, a structural bias guides self-attention toward biologically related tokens, and a graph-aware router uses neighborhood information to determine each token's recursion depth. These techniques are centered on our insight that additional knowledge of token interaction can effectively help models construct embeddings and select which tokens should be learned more deeply. Across eight benchmarks spanning diverse omics data types and evaluated under a unified five-fold cross-validation protocol, bioMoR improves average macro-F1 by 8.2 percentage points and balanced accuracy by 7.1 percentage points over the strongest biology-agnostic MoR baseline while using 75 percent fewer parameters and up to 58 percent fewer FLOPs than a non-recursive Transformer. The selected marker genes or pathways provide biological interpretability, while their token-specific recursion depths reveal how computation is allocated.

Toward Uncertainty Quantification in Modern Art

Asked to animate the same modern artwork under different random seeds, a text to video model returns visibly different films, one reading per seed. Because modern art is ambiguous by intent, this disagreement is signal, not noise. Yet prevailing uncertainty quantification (UQ) collapses a set of generations to a dispersion scalar that says how much the seeds differ but not how: it cannot tell a compact interpretation from a dominant reading plus an outlier, two competing modes, or diffuse instability, nor whether the set still contains a rendering faithful to the original. We present the first study of the structure of generative uncertainty for modern art animation, and a reusable protocol for identifying source blind multiseed uncertainty: a suite of seven source blind and six reference aware estimators; a distributional profile (robust spread, outlier influence, explicit topology, multimodality, anisotropy, leave one seed influence, reference coverage); a distribution model ablation (vMF, Kent, ACG, Student t, kernel, mixture); eight identification questions; and an artwork level statistical protocol. We build the first corpus: 250 modern artwork captions rendered by Wan2.1 14B under four seeds (1000 videos) across 4 encoders, artworks withheld from generation. As a diagnostic the protocol succeeds: it classifies seed set topology at balanced accuracy 0.98 (chance 0.25), isolates the outlier configuration at AUROC 1.00 where a scalar reaches only 0.35, and splits high uncertainty artworks into reference covering (n=97) and reference missing (n=56) diversity, reliably from three seeds and across encoders.

TRAPS: Therapeutic Response Analysis via Pathway-informed Stratification

Cancer treatment planning requires decisions across multiple clinical dimensions at once. Clinicians must determine whether a patient should receive targeted molecular therapy, radiation therapy, and whether they are likely to survive beyond six months. Existing pathway-informed deep learning models have been developed and tested in isolation, making fair comparison across architectures impossible. We present the first unified benchmark for pathway-guided therapy response modeling, evaluating three biologically informed architectures, BINN, GraphPath, and PATH, across five cancer cohorts drawn from The Cancer Genome Atlas, representing 2,622 patients encoded using Reactome pathway activity scores. Each model is trained jointly on all three clinical outcomes under identical data and evaluation conditions, the first study to treat pathway-structured deep learning as a combined therapy and survival prediction problem. Our results show that no single architecture wins across all tasks: PATH performs best for targeted molecular therapy prediction overall, BINN is most reliable for survival prediction, and no model produces useful predictions for radiation therapy, as the key drivers of that decision are clinical variables not captured in gene expression data. Most strikingly, GraphPath achieves an AUROC of 0.92 on prostate targeted molecular therapy prediction, the highest score in the entire benchmark, demonstrating that lateral co-regulation structure produces exceptional discriminative power when matched to a cohort with a narrow targetable driver programme, even under conditions of extreme class imbalance at only 11\% positive prevalence.

Improving the Safety and Trustworthiness of Medical AI via Multi-Agent Evaluation Loops

Large Language Models (LLMs) are increasingly applied in healthcare, yet ensuring their ethical integrity and safety compliance remains a major barrier to clinical deployment. This work introduces a multi-agent refinement framework designed to enhance the safety and reliability of medical LLMs through structured, iterative alignment. Our system combines two generative models - DeepSeek R1 and Med-PaLM - with two evaluation agents, LLaMA 3.1 and Phi-4, which assess responses using the American Medical Association's (AMA) Principles of Medical Ethics and a five-tier Safety Risk Assessment (SRA-5) protocol. We evaluate performance across 900 clinically diverse queries spanning nine ethical domains, measuring convergence efficiency, ethical violation reduction, and domain-specific risk behavior. Results demonstrate that DeepSeek R1 achieves faster convergence (mean 2.34 vs. 2.67 iterations), while Med-PaLM shows superior handling of privacy-sensitive scenarios. The iterative multi-agent loop achieved an 89% reduction in ethical violations and a 92% risk downgrade rate, underscoring the effectiveness of our approach. This study presents a scalable, regulator-aligned, and cost-efficient paradigm for governing medical AI safety.

AgEval: A Benchmark for Zero-Shot and Few-Shot Plant Stress Phenotyping with Multimodal LLMs

Plant stress phenotyping traditionally relies on expert assessments and specialized models, limiting scalability in agriculture. Recent advances in multimodal large language models (LLMs) offer potential solutions to this challenge. We present AgEval, a benchmark comprising 12 diverse plant stress phenotyping tasks, to evaluate these models' capabilities. Our study assesses zero-shot and few-shot in-context learning performance of state-of-the-art models, including Claude, GPT, Gemini, and LLaVA. Results show significant performance improvements with few-shot learning, with F1 scores increasing from 46.24% to 73.37% in 8-shot identification for the best-performing model. Few-shot examples from other classes in the dataset have negligible or negative impacts, although having the exact category example helps to increase performance by 15.38%. We also quantify the consistency of model performance across different classes within each task, finding that the coefficient of variance (CV) ranges from 26.02% to 58.03% across models, implying that subject matter expertise is needed - of 'difficult' classes - to achieve reliability in performance. AgEval establishes baseline metrics for multimodal LLMs in agricultural applications, offering insights into their promise for enhancing plant stress phenotyping at scale. Benchmark and code can be accessed at: https://anonymous.4open.science/r/AgEval/