Anirudh Thatipelli

Ali Siahkoohi, Anirudh Thatipelli

Input-convex neural networks (ICNNs) are widely used for log-concave density estimation, convex-potential normalizing flows, optimal transport, and transport-map inversion for high-dimensional Bayesian posteriors. These tasks share a structural constraint: the inter-layer weights of the ICNN must remain non-negative. The standard recipe, projected gradient descent (PGD) onto the non-negative cone, applies a hard, non-smooth projection -- the stiff-penalty limit of an ADMM-style constraint splitting -- and its classical convergence guarantees do not transfer to the non-smooth ICNN training landscape; the differentiable alternative, softplus reparametrization, attenuates the gradient exponentially in the weight magnitude, stalling training with dead inter-layer weights and plateaued loss. Inspired by parameter-extension lifts of PDE-constrained inverse problems, we propose the lift: instead of constraining the inter-layer weights directly, we train an unconstrained hypernetwork that emits them from a permutation-invariant summary of the input batch. This adds stochasticity to the training dynamics that softens the loss landscape, letting the iterates escape the gradient-attenuated region where direct softplus stalls. We trace this softening to three structural ingredients -- a learnable bias acting as slack, a hypernetwork body that conditions on the target batch, and a cross-covariance coupling the two through batch stochasticity -- and prove each one necessary: deleting any single ingredient collapses the cross-covariance that carries the softening. On log-concave energy-based modeling from one-dimensional toy targets to image-flavored latents, and convex-potential normalizing flows on a 21-dimensional tabular benchmark, we show that the lift reaches a lower test loss than both PGD and direct softplus, and turns a plateau-bounded training trajectory into a valley-descending one.

Anirudh Thatipelli, Sanath Narayan, Salman Khan et al.

We propose a novel few-shot action recognition framework, STRM, which enhances class-specific feature discriminability while simultaneously learning higher-order temporal representations. The focus of our approach is a novel spatio-temporal enrichment module that aggregates spatial and temporal contexts with dedicated local patch-level and global frame-level feature enrichment sub-modules. Local patch-level enrichment captures the appearance-based characteristics of actions. On the other hand, global frame-level enrichment explicitly encodes the broad temporal context, thereby capturing the relevant object features over time. The resulting spatio-temporally enriched representations are then utilized to learn the relational matching between query and support action sub-sequences. We further introduce a query-class similarity classifier on the patch-level enriched features to enhance class-specific feature discriminability by reinforcing the feature learning at different stages in the proposed framework. Experiments are performed on four few-shot action recognition benchmarks: Kinetics, SSv2, HMDB51 and UCF101. Our extensive ablation study reveals the benefits of the proposed contributions. Furthermore, our approach sets a new state-of-the-art on all four benchmarks. On the challenging SSv2 benchmark, our approach achieves an absolute gain of $3.5\%$ in classification accuracy, as compared to the best existing method in the literature. Our code and models are available at https://github.com/Anirudh257/strm.

Neel Trivedi, Anirudh Thatipelli, Ravi Kiran Sarvadevabhatla

The lack of fine-grained joints (facial joints, hand fingers) is a fundamental performance bottleneck for state of the art skeleton action recognition models. Despite this bottleneck, community's efforts seem to be invested only in coming up with novel architectures. To specifically address this bottleneck, we introduce two new pose based human action datasets - NTU60-X and NTU120-X. Our datasets extend the largest existing action recognition dataset, NTU-RGBD. In addition to the 25 body joints for each skeleton as in NTU-RGBD, NTU60-X and NTU120-X dataset includes finger and facial joints, enabling a richer skeleton representation. We appropriately modify the state of the art approaches to enable training using the introduced datasets. Our results demonstrate the effectiveness of these NTU-X datasets in overcoming the aforementioned bottleneck and improve state of the art performance, overall and on previously worst performing action categories. Code and pretrained models can be found at https://github.com/skelemoa/ntu-x .

Anirudh Thatipelli, Ali Siahkoohi

Functional data clustering is concerned with grouping functions that share similar structure, yet most existing methods implicitly operate on sampled grids, causing cluster assignments to depend on resolution, sampling density, or preprocessing choices rather than on the underlying functions themselves. To address this limitation, we introduce a framework that maps discretized function observations -- at arbitrary resolution and on arbitrary grids -- into a fixed-dimensional vector space via an auto-encoding architecture. The encoder is a hypernetwork that maps coordinate-value pairs to the weight space of an implicit neural representation (INR), which serves as the decoder. Because INRs represent functions with very few parameters, this design yields compact representations that are decoupled from the sampling grid, while the hypernetwork amortizes weight prediction across the dataset. Clustering is then performed in this weight space using standard algorithms, making the approach agnostic to both the discretization and the choice of clustering method. By means of synthetic and real-world experiments in high-dimensional settings, we demonstrate competitive clustering performance that is robust to changes in sampling resolution -- including generalization to resolutions not seen during training.

Anirudh Thatipelli, Shao-Yuan Lo, Amit K. Roy-Chowdhury

Egocentric vision captures the scene from the point of view of the camera wearer, while exocentric vision captures the overall scene context. Jointly modeling ego and exo views is crucial to developing next-generation AI agents. The community has regained interest in the field of egocentric vision. While the third-person view and first-person have been thoroughly investigated, very few works aim to study both synchronously. Exocentric videos contain many relevant signals that are transferrable to egocentric videos. This paper provides a timely overview of works combining egocentric and exocentric visions, a very new but promising research topic. We describe in detail the datasets and present a survey of the key applications of ego-exo joint learning, where we identify the most recent advances. With the presentation of the current status of the progress, we believe this short but timely survey will be valuable to the broad video-understanding community, particularly when multi-view modeling is critical.

Kevin Zhai, Utsav Singh, Anirudh Thatipelli et al.

Diffusion models excel at generating images conditioned on text prompts, but the resulting images often do not satisfy user-specific criteria measured by scalar rewards such as Aesthetic Scores. This alignment typically requires fine-tuning, which is computationally demanding. Recently, inference-time alignment via noise optimization has emerged as an efficient alternative, modifying initial input noise to steer the diffusion denoising process towards generating high-reward images. However, this approach suffers from reward hacking, where the model produces images that score highly, yet deviate significantly from the original prompt. We show that noise-space regularization is insufficient and that preventing reward hacking requires an explicit image-space constraint. To this end, we propose MIRA (MItigating Reward hAcking), a training-free, inference-time alignment method. MIRA introduces an image-space, score-based KL surrogate that regularizes the sampling trajectory with a frozen backbone, constraining the output distribution so reward can increase without off-distribution drift (reward hacking). We derive a tractable approximation to KL using diffusion scores. Across SDv1.5 and SDXL, multiple rewards (Aesthetic, HPSv2, PickScore), and public datasets (e.g., Animal-Animal, HPDv2), MIRA achieves >60\% win rate vs. strong baselines while preserving prompt adherence; mechanism plots show reward gains with near-zero drift, whereas DNO drifts as compute increases. We further introduce MIRA-DPO, mapping preference optimization to inference time with a frozen backbone, extending MIRA to non-differentiable rewards without fine-tuning.

Pranay Gupta, Anirudh Thatipelli, Aditya Aggarwal et al.

In this paper, we study current and upcoming frontiers across the landscape of skeleton-based human action recognition. To study skeleton-action recognition in the wild, we introduce Skeletics-152, a curated and 3-D pose-annotated subset of RGB videos sourced from Kinetics-700, a large-scale action dataset. We extend our study to include out-of-context actions by introducing Skeleton-Mimetics, a dataset derived from the recently introduced Mimetics dataset. We also introduce Metaphorics, a dataset with caption-style annotated YouTube videos of the popular social game Dumb Charades and interpretative dance performances. We benchmark state-of-the-art models on the NTU-120 dataset and provide multi-layered assessment of the results. The results from benchmarking the top performers of NTU-120 on the newly introduced datasets reveal the challenges and domain gap induced by actions in the wild. Overall, our work characterizes the strengths and limitations of existing approaches and datasets. Via the introduced datasets, our work enables new frontiers for human action recognition.