Vyom Sharma

Ali Abedi, Paritosh Bisht, Riddhi Chatterjee et al.

Physical exercise is an essential component of rehabilitation programs that improve quality of life and reduce mortality and re-hospitalization rates. In AI-driven virtual rehabilitation programs, patients complete their exercises independently at home, while AI algorithms analyze the exercise data to provide feedback to patients and report their progress to clinicians. To analyze exercise data, the first step is to segment it into consecutive repetitions. There has been a significant amount of research performed on segmenting and counting the repetitive activities of healthy individuals using raw video data, which raises concerns regarding privacy and is computationally intensive. Previous research on patients' rehabilitation exercise segmentation relied on data collected by multiple wearable sensors, which are difficult to use at home by rehabilitation patients. Compared to healthy individuals, segmenting and counting exercise repetitions in patients is more challenging because of the irregular repetition duration and the variation between repetitions. This paper presents a novel approach for segmenting and counting the repetitions of rehabilitation exercises performed by patients, based on their skeletal body joints. Skeletal body joints can be acquired through depth cameras or computer vision techniques applied to RGB videos of patients. Various sequential neural networks are designed to analyze the sequences of skeletal body joints and perform repetition segmentation and counting. Extensive experiments on three publicly available rehabilitation exercise datasets, KIMORE, UI-PRMD, and IntelliRehabDS, demonstrate the superiority of the proposed method compared to previous methods. The proposed method enables accurate exercise analysis while preserving privacy, facilitating the effective delivery of virtual rehabilitation programs.

Debajyoti Datta, Trishala Neeraj, Bibek Paudel et al.

Long-context inference is constrained by KV-cache memory, which grows linearly with sequence length; KV-cache compression therefore hinges on reliably selecting which past tokens to retain. Most geometry-based eviction methods score keys by cosine similarity to a global centroid, but cosine is scale-invariant and can discard magnitude cues that distinguish semantically salient tokens. We propose ManifoldKV, a training-free scorer that ranks tokens by Euclidean distance to the key centroid, capturing both angular and radial deviations. On the RULER benchmark, ManifoldKV achieves 95.7% accuracy at 4K-16K contexts with 20% compression; matching the best geometric baseline while improving robustness in two regimes where cosine scoring fails. First, on multi-key retrieval, ManifoldKV reduces directional collisions, achieving 92.4% vs KeyDiff's 77.0% (+15.4 points) on 3-key NIAH at 50% compression. Second, to address dilution and performance collapse of global centroids at 64K context, we introduce WindowedManifoldKV, which restores accuracy to 84.3% at 25% compression, a 49-point recovery over global L2 and +3.2 points over KeyDiff. The method requires only 3 lines of code and works across 4 architectures without tuning.

Vyom Sharma, Debajyoti Datta

The optimal kernel configuration for Mixture-of-Experts (MoE) inference depends on both batch size and the expert routing distribution, yet production systems dispatch from batch size alone, leaving 10-70% of kernel throughput unrealized. We present RaMP, a routing-aware dispatch framework. A performance-region analysis derives, from hardware constants alone, when each optimization helps, correctly predicting all 8 tested architectures, including 3 unseen. A four-parameter wave cost model selects the fastest configuration from the runtime expert histogram, achieving 0.93% mean regret versus exhaustive search, fitted from just 10-24 minutes of one-time profiling per model. Because the model depends only on CTA grid geometry, it is kernel-agnostic: applied to Alpha-MoE, it delivers 1.14x with no source modification. Paired with a co-designed CuTe DSL kernel exposing 134-268 polymorphic configurations, RaMP delivers 1.22x kernel speedup over static dispatch and 1.30x end-to-end speedup in vLLM serving over Triton, 1.41x over DeepGEMM, and 1.13x over FlashInfer CUTLASS.