Akash Yadav

Akash Yadav

PhD Candidate

Posts by Tags

ai

NeurIPS 2025: Moments and Takeaways from a Landmark Conference

2 minute read

Published:

Attending NeurIPS 2025 was an unforgettable experience that left me both inspired and contemplative about the future of artificial intelligence. It was my first time at the conference, and I was struck by the sheer scale and diversity of the event. From cutting-edge research presentations to lively panel discussions and networking opportunities, NeurIPS felt like a microcosm of the broader AI ecosystem. One of the most striking aspects of NeurIPS 2025 was the palpable sense of momentum in the field. Advances in large language models, multimodal systems, and reinforcement learning were on full display, with researchers showcasing impressive results that pushed the boundaries of what AI can achieve. At the same time, there was a strong emphasis on ethical considerations, safety, and societal impact, reflecting a growing awareness of the responsibilities that come with developing powerful technologies.

calibration

Stochastic Attention for Uncertainty-Aware Scientific Foundation Models

3 minute read

Published:

Scientific foundation models are increasingly being used for forecasting, surrogate modeling, and large-scale scientific prediction. They are powerful because they can reuse representations across tasks and domains, but most of them still behave deterministically at inference time. For high-stakes scientific settings, that is a serious limitation.

conference

NeurIPS 2025: Moments and Takeaways from a Landmark Conference

2 minute read

Published:

Attending NeurIPS 2025 was an unforgettable experience that left me both inspired and contemplative about the future of artificial intelligence. It was my first time at the conference, and I was struck by the sheer scale and diversity of the event. From cutting-edge research presentations to lively panel discussions and networking opportunities, NeurIPS felt like a microcosm of the broader AI ecosystem. One of the most striking aspects of NeurIPS 2025 was the palpable sense of momentum in the field. Advances in large language models, multimodal systems, and reinforcement learning were on full display, with researchers showcasing impressive results that pushed the boundaries of what AI can achieve. At the same time, there was a strong emphasis on ethical considerations, safety, and societal impact, reflecting a growing awareness of the responsibilities that come with developing powerful technologies.

forecasting

Stochastic Attention for Uncertainty-Aware Scientific Foundation Models

3 minute read

Published:

Scientific foundation models are increasingly being used for forecasting, surrogate modeling, and large-scale scientific prediction. They are powerful because they can reuse representations across tasks and domains, but most of them still behave deterministically at inference time. For high-stakes scientific settings, that is a serious limitation.

neurips

NeurIPS 2025: Moments and Takeaways from a Landmark Conference

2 minute read

Published:

Attending NeurIPS 2025 was an unforgettable experience that left me both inspired and contemplative about the future of artificial intelligence. It was my first time at the conference, and I was struck by the sheer scale and diversity of the event. From cutting-edge research presentations to lively panel discussions and networking opportunities, NeurIPS felt like a microcosm of the broader AI ecosystem. One of the most striking aspects of NeurIPS 2025 was the palpable sense of momentum in the field. Advances in large language models, multimodal systems, and reinforcement learning were on full display, with researchers showcasing impressive results that pushed the boundaries of what AI can achieve. At the same time, there was a strong emphasis on ethical considerations, safety, and societal impact, reflecting a growing awareness of the responsibilities that come with developing powerful technologies.

reflections

NeurIPS 2025: Moments and Takeaways from a Landmark Conference

2 minute read

Published:

Attending NeurIPS 2025 was an unforgettable experience that left me both inspired and contemplative about the future of artificial intelligence. It was my first time at the conference, and I was struck by the sheer scale and diversity of the event. From cutting-edge research presentations to lively panel discussions and networking opportunities, NeurIPS felt like a microcosm of the broader AI ecosystem. One of the most striking aspects of NeurIPS 2025 was the palpable sense of momentum in the field. Advances in large language models, multimodal systems, and reinforcement learning were on full display, with researchers showcasing impressive results that pushed the boundaries of what AI can achieve. At the same time, there was a strong emphasis on ethical considerations, safety, and societal impact, reflecting a growing awareness of the responsibilities that come with developing powerful technologies.

scientific-foundation-models

Stochastic Attention for Uncertainty-Aware Scientific Foundation Models

3 minute read

Published:

Scientific foundation models are increasingly being used for forecasting, surrogate modeling, and large-scale scientific prediction. They are powerful because they can reuse representations across tasks and domains, but most of them still behave deterministically at inference time. For high-stakes scientific settings, that is a serious limitation.

stochastic-attention

Stochastic Attention for Uncertainty-Aware Scientific Foundation Models

3 minute read

Published:

Scientific foundation models are increasingly being used for forecasting, surrogate modeling, and large-scale scientific prediction. They are powerful because they can reuse representations across tasks and domains, but most of them still behave deterministically at inference time. For high-stakes scientific settings, that is a serious limitation.

uncertainty

Stochastic Attention for Uncertainty-Aware Scientific Foundation Models

3 minute read

Published:

Scientific foundation models are increasingly being used for forecasting, surrogate modeling, and large-scale scientific prediction. They are powerful because they can reuse representations across tasks and domains, but most of them still behave deterministically at inference time. For high-stakes scientific settings, that is a serious limitation.