The capabilities of an LLM agent depend not only on its model but on the harness: the executable program that constructs context, invokes tools, verifies results, and recovers from failure. Recent work shows that evolving the harness yields persistent improvements without updating model weights. Existing approaches, however, assume that all execution experience can be routed to a single optimizer, which evolves one harness along a sequential trajectory. Real agent ecosystems violate that assumption: users, organizations, and environments generate isolated streams of experience that cannot be pooled, so the experience most worth learning from is exactly the experience that cannot be directly centralized. We introduce EvolveNet, a paradigm of collaborative harness evolution that moves experience extraction to the data. A shared harness is broadcast to data-local agent deployments, each of which evolves it on its own workload. Only the resulting program adaptations are composed into an updated shared harness and redistributed, so that every participating agent inherits operational experience discovered by the others. By shifting the aggregation boundary from raw workloads to learned adaptations, EvolveNet keeps workloads local and allows multiple evolutionary searches to proceed concurrently with reduced serial depth. Because independently modified programs cannot be averaged like model parameters and may conflict when composed, EvolveNet introduces scope-typed, evidence-guided program aggregation. Across five settings spanning text-to-SQL, data-science coding, competitive programming, software engineering, and agentic workflows, EvolveNet improves the shared harness in all five, with the largest gains under heterogeneous workloads, and ablations attribute the improvement to composition of adaptations from different agents rather than to selecting among them.
Jun Nie, Yonggang Zhang, Qi Cai et al.· 1 citation
Detecting and rejecting out-of-distribution (OOD) data is crucial for enhancing the reliability and minimizing potential risks associated with models, such as neural networks, in their deployment phase. In this paper, we find that a well-pretrained in-distribution (ID) model can memorize and recognize ID patterns, even when the features undergo alterations. The networks remain robust on the altered ID features whereas the OOD features are heavily distorted, containing distinctive clues for OOD detection in the feature space. Therefore, we introduce a novel method, Feature-alteration Robustness (FAR), designed to efficiently detect OOD samples by measuring feature map robustness under alterations. Specifically, FAR alters the feature maps of intermediate layers, and then evaluates the foreground-background deviations after several layers. We provide a theoretical analysis to help understand our method FAR. The experimental results show that our methods FAR and FAR+ASH can achieve state-of-the-art on various benchmarks.
Xue Jiang, Feng Liu, Zhen Fang et al.· IEEE Transactions on Pattern...· 0 citations
Deep research agents increasingly operate over the open web, where relevant records coexist with redundant summaries, outdated reports, and misleading documents. Existing evaluations offer limited insight into whether agents preserve sound evidential standards when an ordinary-looking false document is deliberately seeded into a searchable environment and offers a direct shortcut to a conflicting answer. We introduce DRNOISE, a 100-task benchmark for answer recovery under misleading evidence. Each task has a unique gold answer supported by two corroborating indirect record chains; the paired noisy condition adds one plausible document that states a conflicting answer directly. The benchmark spans ten families of evidence operations. Across agents with strong clean-task performance, this single intervention causes 66-88 percentage-point accuracy drops. Trace analyses identify verification inertia as the dominant failure mode: agents often retrieve truthful records but stop before completing and reconciling the evidence chain, instead deferring to the answer-like document. Generic verification prompts reduce but do not close this gap. The setting is especially relevant to open-web deployment, where plausible falsehoods arrive through ordinary-looking pages rather than explicit attacks. Reliable deep research therefore requires more than retrieval and citation; it requires active reconciliation of direct claims with record-level evidence.
Jun Nie, Zhiqin Yang, Zhenheng Tang et al.· 1 citation
This work establishes a new paradigm for generated image detection by recasting the detection task as a problem of machine unlearning, and introduces two detection methods: data-free detection, which prunes model parameters to induce unlearning without data access, and data-driven detection, which optimizes LVMs to unlearn knowledge tied to generated images.
Jun Nie, Yonggang Zhang, Tongliang Liu et al.· 0 citations
Test-Time Harness Evolution is introduced, which treats the executable harness as the state of test-time adaptation for LLM agents as evolution over executable control programs and identifies execution-derived proxy reliability as a central challenge for robust unsupervised agent improvement.
Jun Nie, Yonggang Zhang, Jun Song et al.· 3 citations