Bharath Kumar Shanmugam
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
Bharath Kumar Shanmugam has a total of 73 co-authors. Here are the top co-authors by number of shared publications:
Harsha Vardhan
National Institute of Technology Karnataka
17 shared publicationsMarutiram Kaza
Research &Development Center, JSW Steel Limited, Vijayanagar Works, Ballari, India
17 shared publicationsRameshwar Sah
Research &Development Center, JSW Steel Limited, Vijayanagar Works, Ballari, India
17 shared publicationsHarish Hanumanthappa
RV University
15 shared publicationsM. Govinda Raj
National Institute of Technology Karnataka
15 shared publications- Harish Hanumanthappa
School of Computer Science and Engineering, RV University, Bangalore, India
7 shared publications
Latest Publications
- Open Access
- 5 October 2026
- Published by Springer Nature in Research Square
When Validating the Judge Changes the Result: Instability in Human Validation of LLM Safety Graders
Safety verdicts from an automated judge built on a large language model are usually checked against human labels before anyone trusts them. We show that this check can return very different answers depending on how it is run. We study one grader, one rubric and one corpus: 300 matched prompt pairs over 25 harmful intents, each written once in formal benchmark style and once in casual first-person style, answered by six instruction- tuned models. A keyword refusal metric finds a casual-minus-formal gap of 0.263. For actionable harm, the answer depends on the validation. One unblinded annotator implies 0.258. Two unblinded annotators given the same 442 items imply−0.036 and +0.274, agree at Cohen’s κ = 0.09, and split 95 times on whether the model refused at all. A pre-registered, blinded audit with three label sets implies 0.035. Regenerating all 3,600 responses without a length cap and grading them at five budgets moved neither automated gap. Across instruments, the gap shrinks from 0.263 to 0.035 as the decision boundary moves from refusal to actionability, with each instrument close to what its own boundary predicts. Neither factor explains why the human audits disagree. What is left is the val- idation procedure, although we cannot say which part of it is responsible. On the same responses, the protocol alone decides whether a model passes a release threshold. We argue that validating a judge is a measurement experiment in its own right, to be specified, blinded and reported per class, and propose a reporting checklist.
- 5 October 2026
- Published by Springer Nature in Research Square
Open AccessWhen Validating the Judge Changes the Result: Instability in Human Validation of LLM Safety Graders
Safety verdicts from an automated judge built on a large language model are usually checked against human labels before anyone trusts them. We show that this check can return very different answers depending on how it is run. We study one grader, one rubric and one corpus: 300 matched prompt pairs over 25 harmful intents, each written once in formal benchmark style and once in casual first-person style, answered by six instruction- tuned models. A keyword refusal metric finds a casual-minus-formal gap of 0.263. For actionable harm, the answer depends on the validation. One unblinded annotator implies 0.258. Two unblinded annotators given the same 442 items imply−0.036 and +0.274, agree at Cohen’s κ = 0.09, and split 95 times on whether the model refused at all. A pre-registered, blinded audit with three label sets implies 0.035. Regenerating all 3,600 responses without a length cap and grading them at five budgets moved neither automated gap. Across instruments, the gap shrinks from 0.263 to 0.035 as the decision boundary moves from refusal to actionability, with each instrument close to what its own boundary predicts. Neither factor explains why the human audits disagree. What is left is the val- idation procedure, although we cannot say which part of it is responsible. On the same responses, the protocol alone decides whether a model passes a release threshold. We argue that validating a judge is a measurement experiment in its own right, to be specified, blinded and reported per class, and propose a reporting checklist.
- Open Access
- 5 October 2026
- Published by Springer Nature in Research Square
Position, not importance: what a likelihood threshold actually defers on
Empower, a recent assistant-training method, teaches a language model to stop writing where its continuation stops being predictable, assuming predictable means unimportant. We test this on code. CEDE-Bench has 124 hand-authored items: 112 points where a Python completion chooses a dangerous or safe branch, and 12 benign controls with no consequential choice. At η = 4, the user-study threshold, its stopping rule writes the dangerous branch on 37 of 112 decisions (33.0%, Wilson 95% confidence interval 25.0 to 42.2) and on 8 of 10 worst-case probes. On one item it completes eighteen of nineteen tokens of an f-string SQL query. At the same threshold it cuts into 11 of 12 benign controls, finishing about a third of boilerplate. Its stopping point is statistically indistinguishable between consequential and benign items (mean completed fraction 0.26 against 0.35, Mann–Whitney p = 0.16). Position decides it. The commit rate correlates with the decision token's index at r = −0.56 and with its probability at r = +0.26; it is 76% for decisions in the first two tokens and 0 of 35 past token ten. Under a roughly constant entropy rate, the boundary reduces to η ln 2/H, a token count set by local predictability; this recovers the observed stop with R² = 0.55 and a median error of one token. At η = 0.32 the commit rate falls to 8.9%, but only 7.5% of benign text gets completed, so no point on the sweep is both safe and useful. Data and code are released.
- 5 October 2026
- Published by Springer Nature in Research Square
Open AccessPosition, not importance: what a likelihood threshold actually defers on
Empower, a recent assistant-training method, teaches a language model to stop writing where its continuation stops being predictable, assuming predictable means unimportant. We test this on code. CEDE-Bench has 124 hand-authored items: 112 points where a Python completion chooses a dangerous or safe branch, and 12 benign controls with no consequential choice. At η = 4, the user-study threshold, its stopping rule writes the dangerous branch on 37 of 112 decisions (33.0%, Wilson 95% confidence interval 25.0 to 42.2) and on 8 of 10 worst-case probes. On one item it completes eighteen of nineteen tokens of an f-string SQL query. At the same threshold it cuts into 11 of 12 benign controls, finishing about a third of boilerplate. Its stopping point is statistically indistinguishable between consequential and benign items (mean completed fraction 0.26 against 0.35, Mann–Whitney p = 0.16). Position decides it. The commit rate correlates with the decision token's index at r = −0.56 and with its probability at r = +0.26; it is 76% for decisions in the first two tokens and 0 of 35 past token ten. Under a roughly constant entropy rate, the boundary reduces to η ln 2/H, a token count set by local predictability; this recovers the observed stop with R² = 0.55 and a median error of one token. At η = 0.32 the commit rate falls to 8.9%, but only 7.5% of benign text gets completed, so no point on the sweep is both safe and useful. Data and code are released.
- 31 August 2026
- Published by Springer Nature in Mobile Networks and Applications
- Vol. 31 (3-4) , 347-362
Machine Learning - Based Classification Framework for Intelligent 5G Resource Allocation and Network Optimization
- 31 August 2026
- Published by Springer Nature in Mobile Networks and Applications
- Vol. 31 (3-4) , 347-362
Machine Learning - Based Classification Framework for Intelligent 5G Resource Allocation and Network Optimization
- 21 February 2025
- Published by World Scientific Pub Co Pte Ltd in Surface Review and Letters
COMPARISON OF THE PERFORMANCE OF INDIRECT EVALUATION OF FLANK WEAR FOR TURNING INCONEL-718 USING THE PROCESSED IMAGE AND ACOUSTIC WAVES
This study investigates the tool wear for turning Inconel-718 using the titanium nitride carbide tool inserts. This research work aims to compare the performance of processed images and acoustic waves used as an indirect technique for evaluating tool wear. The work targets to capture the tool wear and tool image after turning and acoustic wave during turning for each experimental run. The pixel area of the processed picture, the root-mean-square (RMS) of the acoustic wave, and the microscope tool wear of the tool maker were taken into consideration as output parameters for the change of operational parameters including feed, speed, and depth of cut. The performance of wear, pixel area, and RMS was compared using the Box Behnken method. Further, the correlation between the performance of tool wear, image processed pixel area, and RMS for the variation in input variable was obtained from interaction and main effects plots. The results demonstrated that at lower speeds (280[ ]rpm), lower feed rates (0.04[ ]mm/rev), and medium depth of cut (0.2[ ]mm), there was less wear, pixel area, and RMS. Wear, pixel area, and RMS have all decreased as a result of the tool and workpiece having less surface friction due to the reduced speed, feed, and medium depth of cut. From the analysis, it was also clear that the indirect evaluation of the wear can be successfully carried out using digital image pixel area and acoustic wave RMS for turning Inconel-718 using a titanium nitride-coated carbide tool.
- 21 February 2025
- Published by World Scientific Pub Co Pte Ltd in Surface Review and Letters
COMPARISON OF THE PERFORMANCE OF INDIRECT EVALUATION OF FLANK WEAR FOR TURNING INCONEL-718 USING THE PROCESSED IMAGE AND ACOUSTIC WAVES
This study investigates the tool wear for turning Inconel-718 using the titanium nitride carbide tool inserts. This research work aims to compare the performance of processed images and acoustic waves used as an indirect technique for evaluating tool wear. The work targets to capture the tool wear and tool image after turning and acoustic wave during turning for each experimental run. The pixel area of the processed picture, the root-mean-square (RMS) of the acoustic wave, and the microscope tool wear of the tool maker were taken into consideration as output parameters for the change of operational parameters including feed, speed, and depth of cut. The performance of wear, pixel area, and RMS was compared using the Box Behnken method. Further, the correlation between the performance of tool wear, image processed pixel area, and RMS for the variation in input variable was obtained from interaction and main effects plots. The results demonstrated that at lower speeds (280[ ]rpm), lower feed rates (0.04[ ]mm/rev), and medium depth of cut (0.2[ ]mm), there was less wear, pixel area, and RMS. Wear, pixel area, and RMS have all decreased as a result of the tool and workpiece having less surface friction due to the reduced speed, feed, and medium depth of cut. From the analysis, it was also clear that the indirect evaluation of the wear can be successfully carried out using digital image pixel area and acoustic wave RMS for turning Inconel-718 using a titanium nitride-coated carbide tool.
- 16 February 2025
- Published by Taylor & Francis in Canadian Metallurgical Quarterly
- Vol. 65 (1) , 935-946
Experimental and prediction analysis of separating blast furnace slag in the vibratory separator
- 16 February 2025
- Published by Taylor & Francis in Canadian Metallurgical Quarterly
- Vol. 65 (1) , 935-946
Experimental and prediction analysis of separating blast furnace slag in the vibratory separator
- 20 October 2023
- Published by Springer Nature in Journal of The Institution of Engineers (India): Series D
- Vol. 105 (3) , 1801-1808
Experimental Analysis of the Rice Husk and Eggshell Powder-Based Natural Fibre Composite
- 20 October 2023
- Published by Springer Nature in Journal of The Institution of Engineers (India): Series D
- Vol. 105 (3) , 1801-1808
Experimental Analysis of the Rice Husk and Eggshell Powder-Based Natural Fibre Composite