Diffusion-model-based image restoration for reliable perception of autonomous bulldozers in complex mining environments

Modeling Environmentally Driven Seasonal Moisture Migration and Ground Movements in Expansive Clays

This paper describes the formulation of a numerical model for simulating environmentally driven one-dimensional (1D) ground movements of expansive clay. The formulation is based on a finite-element model that simulates the redistribution of matric suction through a diffusion-type equation, explicitly accounting for volume changes due to wetting and drying of the clay. We synthesize and modify highly nonlinear constitutive relationships for (1) hysteretic soil water retention; (2) reversible soil shrinkage and expansion of clay; and (3) hydraulic conductivity, explicitly incorporating desiccation cracks through a multidomain framework and assuming a critical surface crack depth. These models are well-calibrated to published laboratory tests on a reference expansive clay, Denver bentonite. We demonstrate capabilities of the proposed formulation to simulate the response of a homogeneous expansive clay to periods of drying and wetting, considering the initial matric suction, saturated hydraulic conductivity of the intact clay, and critical crack depth as three primary sources of uncertainty. We compare ensemble model simulations with measured ground movements from an instrumented expansive clay test site in Texas over a 3-year period using detailed records of potential evapotranspiration and precipitation. By assigning weights to the ensemble simulations based on their performance, we constrain the ranges of the three key uncertain parameters. The results showed very reasonable first-order agreement with the measured data and highlight the potential of the proposed formulation. We anticipate that more reliable predictions can be achieved through direct measurements of actual in situ evaporation rates and local soil properties.

Modelling the impact of temperature on nanocarrier behavior: Thermodynamics, structural transitions, and drug release.

A rational design for next-generation thermo-responsive nanocarriers is proposed, in which polymer chemistry, nanoparticle structure, experimental characterization, and mechanistic modelling are integrated from the earliest stages of material development.

Constitutional Midtraining: Content Presence Drives Alignment Gains

Post-training alignment is often shallow, eroding under fine-tuning. It remains untested as to whether constitutional midtraining interventions can produce durable alignment when cleanly isolated from post-training. We build a 394M-token constitutional corpus from Anthropic's Constitution and apply constitutional midtraining at 120B scale, where principled, values-based content is inserted into midtraining. A 2x2 design (curriculum ordering x deliberative reasoning) was used to produce four constitutionally midtrained conditions, plus a control, which were evaluated on self-generated and established benchmarks including alignment under pressure, value conflict resolution, blackmail, and emergent misalignment. All models were evaluated across three stages: post-midtraining, post-SFT, and post-benign fine-tuning. Constitutionally midtrained models outperformed the control on alignment generalization and durability, notably on blackmail: SFT instilled a blackmail propensity in all models, but constitutional midtraining blunted it, with the advantage surviving benign fine-tuning (-17.5pp). This durability did not extend to settings that required active resistance to in-context pressure or conflict, where the advantage attenuates after SFT. The presence of constitutional content at midtraining also mattered more than its structure, and constitutional midtraining incurred no capability cost, on average, at any stage (MMLU, ARC-Easy, piqa, GSM8K). A modest amount of constitutional content at midtraining could therefore yield broad, persistent alignment gains, offering a cheap, complementary addition to SFT-centered pipelines. Code, data, and models are available.

Flow-accelerated diffusion model for trajectory generation and optimization in offline reinforcement learning

Diffusion-enhanced multi-perspective inconsistency modeling for graph anomaly detection

Measuring Legislature-Aligned Privacy Risks in Synthetic Graphs

SyntheGrAnon is introduced, a framework for evaluating synthetic graph anonymity that primarily targets the singling out, linkability, and inference risks outlined in the EU GDPR at the node and community levels, while also including edge-level attacks as an extension of the node-level setting.