February 2026

A common recommendation for preventing brute-force authentication, credential stuffing, scraping, and resource exhaustion is rate limiting, a standard control for API availability and security. Uncertain semantics under horizontal scaling, proxy-induced client-IP ambiguity, unsafe identity binding, ambiguous policy resolution, canonicalization gaps in endpoint matching, and unbounded in-memory state are some of the reasons why deployments frequently fail. In this paper, we propose an endpoint-aware mechanism that uses token buckets with bounded bucket state to enforce per-endpoint RPS limits that […]

Ver mais

Hi everyone, I’m currently training a small language model to improve its accuracy on code execution prediction (i.e., predicting the exact output from the code and input). I’m working with the Qwen3-4B model and have been using GRPO for training. By combining various dense reward signals, I was able to increase the accuracy to around 72%. This approach also helped eliminate the infinite Repeat Curse(a common problem in smaller Qwen models), and overall training has been stable and […]

Ver mais

Machine learning holds promise for advancing clinical decision support, yet it remains unclear when multimodal learning truly helps in practice, particularly under modality missingness and fairness constraints. In this work, we conduct a systematic benchmark of multimodal fusion between Electronic Health Records (EHR) and chest X-rays (CXR) on standardized cohorts from MIMIC-IV and MIMIC-CXR, aiming to answer four fundamental questions: when multimodal fusion improves clinical prediction, how different fusion strategies compare, how robust existing methods are to missing […]

Ver mais

According to leading management consultancy firm Deloitte, India is slated to corner a big chunk of the data centre growth and expansion happening in the APAC region, which is expected to garner an investment of $800 billion by 2030. Viewed against the challenges faced by the Indian IT sector on account of emergent AI technologies threatening to wipe out traditional avenues of revenue generation, shifting focus to becoming a data centre powerhouse might yet resuscitate the once mighty […]

Ver mais

Continual Learning (CL) requires models to sequentially adapt to new tasks without forgetting old knowledge. Recently, Low-Rank Adaptation (LoRA), a representative Parameter-Efficient Fine-Tuning (PEFT) method, has gained increasing attention in CL. Several LoRA-based CL methods reduce interference across tasks by separating their update spaces, typically building the new space from the estimated null space of past tasks. However, they (i) overlook task-shared directions, which suppresses knowledge transfer, and (ii) fail to capture truly effective task-specific directions since these […]

Ver mais

Hi everyone, I honestly feel like my mental health is not in a good place right now, and I just want to share this to see if anyone else has gone through something similar. If you’ve noticed, I’ve been posting quite a lot recently about my PhD thesis situation. I submitted my thesis a little over two months ago. Since that day, I’ve been in a constant state of anxiety waiting for the result. Every morning, the very […]

Ver mais

Limited interaction between students and lecturers during the learning process often leads to suboptimal information transfer. This issue becomes more critical when addressing multiple integral calculus, a subject that is abstract and requires strong visual comprehension. To address this challenge, innovative learning media that promote student autonomy and support the visualization of complex concepts are needed. This study aims to evaluate the feasibility of a GeoGebra-assisted electronic module (e-module) for a multiple integral calculus course. The research employed […]

Ver mais

Recursive models like TRM/CTM/UT have create a lot of buzz lately. But they’re rarely used outside of static, toy domains – especially language. In 2018, we saw “Universal Transformers” try this. However, follow-up works reveal that simple RLMs (recursive LMs) don’t yield substantial performance gains w.r.t FLOPs spent In this work, I argue that using some rather simple tricks, one can unlock huge performance gains and make RLMs outperform iso-param and iso-FLOP baselines Blogpost/Worklog: https://neel04.github.io/my-website/projects/asura/ Twitter summary thread: […]

Ver mais

The Multiple Traveling Salesman Problem (mTSP) extends the Traveling Salesman Problem to m tours that start and end at a common depot and jointly visit all customers exactly once. In the min-max variant, the objective is to minimize the longest tour, reflecting workload balance. We propose a hybrid approach, Construct, Merge, Solve & Adapt with Reinforcement Learning (RL-CMSA), for the symmetric single-depot min-max mTSP. The method iteratively constructs diverse solutions using probabilistic clustering guided by learned pairwise q-values, […]

Ver mais

In modern cyber-physical vehicle networks, the security of component-level Electronic Control Units (ECUs) is essential for overall system reliability. While CAN bus security is well-studied, the Local Interconnect Network (LIN) has received less attention despite its growing role in critical functions and diagnostic services (UDS). The inherent constraints of the LIN protocol, specifically its low bandwidth and Master-Slave architecture, make traditional fuzz testing impractical due to extremely long execution times. This paper proposes BB-FAST, an optimized framework for […]

Ver mais