Accurate antibody affinity prediction is crucial for drug discovery. However, it remains challenging due to data scarcity, which limits the performance of deep learning models in few-shot, antigen-specific scenarios. To address this challenge, we propose MetaPrior, a novel meta-learning framework designed to learn robust prior knowledge before fine-tuning. The core of MetaPrior is a teacher-student mechanism where a Meta Network (teacher) is trained via bilevel optimization to generate high-quality labels for a diverse set of pseudo-samples and a binding affinity predictor (student) learns from these labels. The teacher is optimized by the student's performance on a small and experimentally measured guide set, enabling the pseudo-samples to approximate the true data distribution under conditions of data scarcity. Extensive experiments demonstrate that MetaPrior is compatible with popular network architectures (e.g., MLPs, CNN s and Transformers) and significantly surpasses standard fine-tuning strategies, particularly in data-scarce scenarios. Across several antigen-specific datasets, such as VEGF, MetaPrior achieves a notable performance gain even with only 40% of the available training data, increasing the Pearson correlation coefficient from 0.38 to 0.58 (a 53% improvement) and enhancing training stability by reducing the standard deviation from 0.42 to 0.19 (a 54.8% reduction in variability). These results confirm the robustness and generalizability of our proposed method. By jointly optimizing pseudo-labels and affinity learning, MetaPrior aligns synthetic supervision with the true data distribution, offering a principled solution for few-shot antibody affinity prediction.

The design of nanobodies with high binding affinity for a given target is the primary objective in the therapeutics development. However, existing generative models often sample the sequence space of nanobody without explicit guidance, which leads to suboptimal affinity. To address this, we introduce DiffNB, a novel diffusion-based framework for controllable, high-affinity nanobody generation. DiffNB is the first diffusion model that leverages Direct Preference Optimization (DPO) to align a pre-trained generative prior with desired bio-physical properties. By fine-tuning the model on preference pairs of high- and low-affinity nanobodies, DiffNB learns to characterize and generate the variants with higher binding affinity and other bio-physical properties improved. During generation, the DPO-aligned DiffNB co-designs the sequence and structure of CDR regions to produce optimized and novel candidates. In our extensive experiments on three therapeutic antigens(HER2, IL-6, and CD45), we demonstrate that DiffNB can generate nanobodies exhibiting higher binding affinity and better structural diversity. Compared to the state-of-the-art baselines, our DPO-guided generation improves the “in silico” binding affinity by up to 35%, structural diversity by up to 46%, stability by up to 8%, and humanness by up to 25%. Our work establishes DPO as a powerful and efficient paradigm for steering generative models in nanobody design, paving the way for targeted and accelerated drug discovery.

Variational inference has experienced a recent surge in popularity owing to stochastic approaches, which have yielded practical tools for a wide range of model classes. A key benefit is that stochastic variational inference obviates the tedious process of deriving analytical expressions for closed-form variable updates. Instead, one simply needs to derive the gradient of the log-posterior, which is often much easier. Yet for certain model classes, the log-posterior itself is difficult to optimize using standard gradient techniques. One such example are random field models, where optimization based on gradient linearization has proven popular, since it speeds up convergence significantly and can avoid poor local optima. In this paper we propose stochastic variational inference with gradient linearization (SVIGL). It is similarly convenient as standard stochastic variational inference - all that is required is a local linearization of the energy gradient. Its benefit over stochastic variational inference with conventional gradient methods is a clear improvement in convergence speed, while yielding comparable or even better variational approximations in terms of KL divergence. We demonstrate the benefits of SVIGL in three applications: Optical flow estimation, Poisson-Gaussian denoising, and 3D surface reconstruction.

Non-local methods exploiting the self-similarity of natural signals have been well studied, for example in image analysis and restoration. Existing approaches, however, rely on k-nearest neighbors (KNN) matching in a fixed feature space. The main hurdle in optimizing this feature space w.r.t. application performance is the non-differentiability of the KNN selection rule. To overcome this, we propose a continuous deterministic relaxation of KNN selection that maintains differentiability w.r.t. pairwise distances, but retains the original KNN as the limit of a temperature parameter approaching zero. To exploit our relaxation, we propose the neural nearest neighbors block (N3 block), a novel non-local processing layer that leverages the principle of self-similarity and can be used as building block in modern neural network architectures. We show its effectiveness for the set reasoning task of correspondence classification as well as for image restoration, including image denoising and single image super-resolution, where we outperform strong convolutional neural network (CNN) baselines and recent non-local models that rely on KNN selection in hand-chosen features spaces.

Lacking realistic ground truth data, image denoising techniques are traditionally evaluated on images corrupted by synthesized i.i.d. Gaussian noise. We aim to obviate this unrealistic setting by developing a methodology for benchmarking denoising techniques on real photographs. We capture pairs of images with different ISO values and appropriately adjusted exposure times, where the nearly noise-free low-ISO image serves as reference. To derive the ground truth, careful post-processing is needed. We correct spatial misalignment, cope with inaccuracies in the exposure parameters through a linear intensity transform based on a novel heteroscedastic Tobit regression model, and remove residual low-frequency bias that stems, e.g., from minor illumination changes. We then capture a novel benchmark dataset, the Darmstadt Noise Dataset (DND), with consumer cameras of differing sensor sizes. One interesting finding is that various recent techniques that perform well on synthetic noise are clearly outperformed by BM3D on photographs with real noise. Our benchmark delineates realistic evaluation scenarios that deviate strongly from those commonly used in the scientific literature.

Many existing approaches for image-to-geometry registration assume that either a textured 3D model or a good initial guess of the 3D pose is available to bootstrap the registration process. In this paper we consider the registration of photographs to 3D models even when no texture information is available. This is very challenging as we cannot rely on texture gradients, and even shading gradients are hard to estimate since the lighting conditions are unknown. To that end, we propose average shading gradients , a rendering technique that estimates the average gradient magnitude over all lighting directions under Lambertian shading. We use this gradient representation as the building block of a registration pipeline based on matching sparse features. To cope with inevitable false matches due to the missing texture information and to increase robustness, the pose of the 3D model is estimated in two stages. Coarse pose hypotheses are first obtained from a single correct match each, subsequently refined using SIFT flow, and finally verified. We apply our algorithm to registering images of real-world objects to untextured 3D meshes of limited accuracy. Moreover, we show that registration can be performed even for paintings despite lacking photo-realism.

Many existing approaches for image-to-geometry registration assume that either a textured 3D model or a good initial guess of the 3D pose is available to bootstrap the registration process. In this paper we consider the registration of photographs to 3D models even when no texture information is available. This is very challenging as we cannot rely on texture gradients, and even shading gradients are hard to estimate since the lighting conditions are unknown. To that end, we propose average shading gradients, a rendering technique that estimates the average gradient magnitude over all lighting directions under Lambertian shading. We use this gradient representation as the building block of a registration pipeline based on matching sparse features. To cope with inevitable false matches due to the missing texture information and to increase robustness, the pose of the 3D model is estimated in two stages. Coarse pose hypotheses are first obtained from a single correct match each, subsequently refined using SIFT flow, and finally verified. We apply our algorithm to registering images of real-world objects to untextured 3D meshes of limited accuracy.