Arka Chakraborty

Arka Chakraborty

Auto-generated profile, based on publicly available information

Arka Chakraborty

Texas State University

Research Focus & Metrics

Academic Career Length

4Years (2023-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
6 March 2023 in Cleaner Logistics and Supply Chain
Verified
Journal Article
Open access
ACArka Chakraborty
Arka Chakraborty
AAAl Amin
Al Amin
MAMd Al Amin
Md Al Amin

4 February 2026 in ACM Computing Surveys
Verified
Journal Article
Open access
Shriyank SomvanshiShriyank Somvanshi
Shriyank Somvanshi
MIMonzurul Islam
Monzurul Islam
ARAmir Rafe
Amir Rafe
ATAnannya Ghosh Tusti
Anannya Ghosh Tusti
ACArka Chakraborty
Arka Chakraborty
SDSubasish Das
Subasish Das
Mechanistic interpretability seeks to reverse-engineer the internal logic of neural networks by uncovering human-understandable circuits, algorithms, and causal structures that drive model behavior. Unlike post hoc explanations that describe what models do, this paradigm focuses on why and how they compute, tracing information flow through neurons, attention heads, and activation pathways. This survey provides a high-level synthesis of the field-highlighting its motivation, conceptual foundations, and methodological taxonomy rather than enumerating individual techniques. We organize mechanistic interpretability across three abstraction layers— neurons , circuits , and algorithms —and three evaluation perspectives: behavioral , counterfactual , and causal . We further discuss representative approaches and toolchains that enable structural analysis of modern AI systems, outlining how mechanistic interpretability bridges theoretical insights with practical transparency. Despite rapid progress, challenges persist in scaling these analyses to frontier models, resolving polysemantic representations, and establishing standardized causal benchmarks. By connecting historical evolution, current methodologies, and emerging research directions, this survey aims to provide an integrative framework for understanding how mechanistic interpretability can support transparency, reliability, and governance in large-scale AI.

10 May 2026 in Case Studies on Transport Policy
Verified
Journal Article
Closed access
ACArka Chakraborty
Arka Chakraborty
SSSawgat Ahmed Shuvo
Sawgat Ahmed Shuvo
SRSharif Ahmed Rafat
Sharif Ahmed Rafat
SDSubasish Das
Subasish Das

3 September 2026 in SSRN Electronic Journal
Verified
Preprint
Open access
ACArka Chakraborty
Arka Chakraborty
ABAnika Baitullah
Anika Baitullah
SDSubasish Das
Subasish Das
Older pedestrians face elevated injury risk in failed-to-yield crashes because crossing decisions often occur under limited visibility, complex vehicle movements, and constrained walking time. This study examines failed-to-yield crashes involving pedestrians aged 65 years and older in Texas from 2021 to 2025. The final dataset included 2,023 crashes, grouped into fatal-and-serious-injury crashes and minor-to-non-injury crashes. A three-stage analytical framework was used. Random Forest (RF) first selected key structured crash variables. Association Rule Mining (ARM) with the Lift Increase Criterion (LIC) then identified recurring crash-factor combinations. Structural Topic Modeling (STM) was applied to police crash narratives to examine whether the structured crash-factor patterns identified through ARM were also reflected in recurring narrative themes. The findings indicate that older pedestrian failed-to-yield crashes are associated with repeated system-level conditions rather than isolated road-user errors. Minor-to-non-injury crashes were more strongly linked with straight-through vehicle movement, passenger cars, city-street contexts, signalized locations, and lower-speed environments. Narrative topics supported the ARM findings by highlighting visibility limitation, failed yielding, signal context, crossing location, turning movement, and vehicle path. The study provided safe system approach (SSA) based policy recommendations by focusing on nighttime crossing visibility, formalized midblock crossings, survivable-speed management, turning-conflict control, age-sensitive signal timing, and vehicle-side pedestrian detection.

24 July 2026 in SSRN Electronic Journal
Verified
Preprint
Open access
ABAnika Baitullah
Anika Baitullah
ACArka Chakraborty
Arka Chakraborty
SSSamia Sahid
Samia Sahid
SDSubasish Das
Subasish Das
This study distinguishes hydroplaning-related crashes from general wet-road crashes and identifies how their fatal and injury (KA) crashes and moderate/minor/no injury (BCO) crash patterns differ. The motivation is that hydroplaning-related crashes are often hidden within broader wet-road records, where narratives may describe rain, wet pavement, skidding, or loss of control without directly identifying hydroplaning. Texas CRIS crash data from 2017 to 2025 were analyzed using a narrative-informed and model-interpretable framework. Structural topic modeling was first applied to crash narratives to separate hydroplaning-related, wet-road, and mixed cases. Machine-learning and deep-learning models were then used to predict hydroplaning-related versus wet-road crashes using structured variables and to identify the factors that differentiated the two groups. Probability-thresholded samples were further analyzed using severity-specific Autoencoder-Based Neurosymbolic (AERIAL) association rules for hydroplaning-related KA crashes, hydroplaning-related BCO crashes, wet-road fatal and injury crashes, and wet-road moderate/minor/no injury crashes. The findings show that hydroplaning-related crashes are more control-loss centered, with fatal and injury crash patterns linked to rural two-lane roads, unsafe speed, skidding, overturning, and fixed-object involvement. Wet-road crashes are more conflict- and recovery-centered, with fatal and injury crash patterns linked to rural high-speed two-lane exposure, no-passing zones, unsafe speed, and opposite-direction conflicts. The findings support separate wet-weather safety strategies, including pavement grooving, drainage improvement, and roadway-departure treatments for hydroplaning-related crashes, speed management, wet-reflective markings, rumble strips, and curve-warning treatments for wet-road crashes.