arXiv subjects

Aryan Sood

Publications and source records attributed to Aryan Sood.

5 recordsLinked to original sources

arXiv2026

Generating Adversarial Texts for Machine Translation via GRPO

As machine translation (MT) systems continue to improve, standard benchmarks become less informative for exposing remaining weaknesses. Traditional methods for creating challenging test sets rely on expensive manual creation or curation, while automated approaches struggle to produce sets with the necessary translation difficulty and linguistic diversity. We propose a scalable reinforcement-learning-based approach for rewriting existing source texts into instances that are more difficult to translate for MT systems. We fine-tune a large language model with Group Relative Policy Optimization (GRPO), using reward signals based on translation difficulty together with constraints for semantic similarity, grammaticality, and approximate length preservation. On WMT25, our approach substantially reduces average COMET translation quality from 0.63 to 0.48, while preserving grammaticality and readability, whereas the base model remains at 0.64. Evaluations on the unseen WMT19-WMT24 benchmarks confirm that this behavior generalizes beyond the training data, and human evaluation further shows that the rewrites substantially lower translation quality while incurring a moderate drop in naturalness and only a small change in grammaticality. We release our code to support reproducibility.

cs.CL↗

arXiv2026

Lost in Interpolation: Why Predictive Feedback Fails in Diffusion Language Models

Soft-masking accelerates the convergence of Masked Diffusion Language Models (MDLMs). Existing formulations build this blend with linear interpolation (LERP) in the raw embedding space, which implicitly treats that space as Euclidean. We analyze the embedding space of MDLMs and find that the mask and predicted-token embeddings maintain a near-constant angle of (\approx 73^\circ) throughout training, while embedding norms remain essentially flat across vocabulary-frequency rank. These indicate a hyperspherical geometry, for which LERP is the wrong interpolation primitive. We introduce Spherical Soft-Masking (S-SM), a drop-in replacement that aggregates the top-(k) predictions with a Fr'echet mean on the hypersphere and blends this mean with the mask direction using spherical linear interpolation (SLERP), then restores the native mask norm. We evaluate S-SM on continued pre-training of a released 169M-parameter MDLM checkpoint across a wide range of inference-time step budgets, SLERP feedback avoids the training degradation that LERP feedback induces and delivers MAUVE gains of up to 2x over the vanilla MDLM baseline and 27.5-56.1% over TopK/LERP at various sampling budgets, alongside consistently lower generative perplexity (16.9-19.6% over the baseline), while leaving output entropy and convergence essentially unchanged.

cs.CL↗

arXiv2026

Dropping the Anchor: Statistical Context Summarization for Distributed Systems via Pulsar Attention

Inference with large language models (LLMs) on long sequences is computationally expensive due to the quadratic complexity of self-attention. Distributed blockwise methods such as Star Attention reduce this cost by sharding context across hosts, but rely on prepending a static, content-blind copy of the first block to every host. We propose Pulsar Attention, which replaces the static anchor with two lightweight, content-aware components: a small attention-sink prefix that stabilizes softmax, and compact cross-block summaries built via a Max-IDF heuristic that selects chunks containing globally rare tokens. This reduces the Phase 1 per-GPU FLOPs by up to 3.3x over Star Attention while retaining an identical KV cache footprint. On RULER with Llama-3.1-8B-Instruct, Pulsar Attention outperforms Star Attention at sequence lengths up to 128K tokens and remains competitive with dense attention across most tasks, with task-dependent absolute gains of up to 4.7% over the dense baseline.

cs.CL↗

arXiv2026

More Than a Quick Glance: Overcoming the Greedy Bias in KV-Cache Compression

While Large Language Models (LLMs) can theoretically support extensive context windows, their actual deployment is constrained by the linear growth of Key-Value (KV) cache memory. Prevailing compression strategies mitigate this through various pruning mechanisms, yet trade-off semantic recall for memory efficiency. In this work, we present LASER-KV (Layer Accumulated Selection with Exact-LSH Recall), a framework designed to test the limits of KV compression under a strict accumulative budgeting policy. We deviate from the standard fixed summary size approach by implementing a block-wise accumulation strategy governed by a protection divisor (n). This allows us to isolate the effects of compression from sliding window artifacts. Our experiments on the Babilong benchmark reveal performance degradation in previous compression methods by 15-30% on various long context tasks. LASER-KV maintains stable performance, achieving superior accuracies by a margin of upto 10% at 128k. These findings challenge the prevailing assumption that attention scores alone are a sufficient proxy for token utility.

cs.AI↗

arXiv2024

CuRIOS-ED: The Technology Demonstrator for the CubeSats for Rapid Infrared and Optical Surveys Mission

The rise of time-domain astronomy including electromagnetic counterparts to gravitational waves, gravitational microlensing, explosive phenomena, and even astrometry with Gaia, are showing the power and need for surveys with high-cadence, large area, and long time baselines to study the transient universe. A constellation of SmallSats or CubeSats providing wide, instantaneous sky coverage down to 21 Vega mag at optical wavelengths would be ideal for addressing this need. We are assembling CuRIOS-ED (CubeSats for Rapid Infrared and Optical Survey--Exploration Demo), an optical telescope payload which will act as a technology demonstrator for a larger constellation of several hundred 16U CubeSats known as CuRIOS. In preparation for CuRIOS, CuRIOS-ED will launch in late 2025 as part of the 12U Starspec InspireSat MVP payload. CuRIOS-ED will be used to demonstrate the StarSpec ADCS pointing capabilities to <1" and to space-qualify a commercial camera package for use on the full CuRIOS payload. The CuRIOS-ED camera system will utilize a Sony IMX455 CMOS detector delivered in an off-the-shelf Atik apx60 package which we modified to be compatible with operations in vacuum as well as the CubeSat form factor, power, and thermal constraints. By qualifying this commercial camera solution, the cost of each CuRIOS satellite will be greatly decreased (~100x) when compared with current space-qualified cameras with IMX455 detectors. We discuss the CuRIOS-ED mission design with an emphasis on the disassembly, repackaging, and testing of the Atik apx60 for space-based missions. Characterization of the apx60's read noise, dark current, patterned noise, and thermal behavior are reported for a range of temperatures (-35 C to 40 C) and exposure times (0.001s to 30 s). Additionally, we comment on preliminary environmental testing results from a successful thermal vacuum test.

astro-ph.IM↗