Subhash Chandra Kantamneni

Mr. Subhash Chandra Kantamneni


Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

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

Representative Publications

View All Publications
19 November 2024 in Entropy
Verified
Journal Article
Open access
PDF file
SKSubhash Chandra Kantamneni
Subhash Chandra Kantamneni
ZLZiming Liu
Ziming Liu
MTMax Tegmark
Max Tegmark
How do transformers model physics? Do transformers model systems with interpretable analytical solutions or do they create an “alien physics” that is difficult for humans to decipher? We have taken a step towards demystifying this larger puzzle by investigating the simple harmonic oscillator (SHO), x¨+2γx˙+ω02x=0, one of the most fundamental systems in physics. Our goal was to identify the methods transformers use to model the SHO, and to do so we hypothesized and evaluated possible methods by analyzing the encoding of these methods’ intermediates. We developed four criteria for the use of a method within the simple test bed of linear regression, where our method was y=wx and our intermediate was w: (1) Can the intermediate be predicted from hidden states? (2) Is the intermediate’s encoding quality correlated with the model performance? (3) Can the majority of variance in hidden states be explained by the intermediate? (4) Can we intervene on hidden states to produce predictable outcomes? Armed with these two correlational (1,2), weak causal (3), and strong causal (4) criteria, we determined that transformers use known numerical methods to model the trajectories of the simple harmonic oscillator, specifically, the matrix exponential method. Our analysis framework can conveniently extend to high-dimensional linear systems and nonlinear systems, which we hope will help reveal the “world model” hidden in transformers.

10 February 2025 in Physical Review E
Verified
Journal Article
Closed access
SKSubhash Chandra Kantamneni
Subhash Chandra Kantamneni
ZLZiming Liu
Ziming Liu
MTMax Tegmark
Max Tegmark
Integrable partial differential equation (PDE) systems are of great interest in natural science, but are exceedingly rare and difficult to discover. To this end, we introduce OptPDE, a numerical approach that Optimizes PDEs' coefficients via automatic differentiation to maximize their number of conserved quantities, n_{CQ}, and thus discover candidates for integrable systems. The algorithm discovers four families of PDEs with at least one nontrivial conserved quantity. We analytically study one new candidate u_{t}=u_{x}^{3}, revealing its interesting properties. It is worth mentioning that our approach aims to discover systems with as many conserved quantities as possible, which does not guarantee integrability. Despite this limitation, our humble effort suggests that numerical algorithms that leverage automatic differentiation are potentially useful for discovering candidates for integrable systems.

23 February 2025 in arXiv
Verified
Preprint
Open access
PDF file
SKSubhash Chandra Kantamneni
Subhash Chandra Kantamneni
JEJoshua Engels
Joshua Engels
SRSenthooran Rajamanoharan
Senthooran Rajamanoharan
MTMax Tegmark
Max Tegmark
NNNeel Nanda
Neel Nanda
Sparse autoencoders (SAEs) are a popular method for interpreting concepts represented in large language model (LLM) activations. However, there is a lack of evidence regarding the validity of their interpretations due to the lack of a ground truth for the concepts used by an LLM, and a growing number of works have presented problems with current SAEs. One alternative source of evidence would be demonstrating that SAEs improve performance on downstream tasks beyond existing baselines. We test this by applying SAEs to the real-world task of LLM activation probing in four regimes: data scarcity, class imbalance, label noise, and covariate shift. Due to the difficulty of detecting concepts in these challenging settings, we hypothesize that SAEs' basis of interpretable, concept-level latents should provide a useful inductive bias. However, although SAEs occasionally perform better than baselines on individual datasets, we are unable to design ensemble methods combining SAEs with baselines that consistently outperform ensemble methods solely using baselines. Additionally, although SAEs initially appear promising for identifying spurious correlations, detecting poor dataset quality, and training multi-token probes, we are able to achieve similar results with simple non-SAE baselines as well. Though we cannot discount SAEs' utility on other tasks, our findings highlight the shortcomings of current SAEs and the need to rigorously evaluate interpretability methods on downstream tasks with strong baselines.