Literature Review

[This review is intended solely for my personal learning] Paper Info arXiv: 2405.03280v2 Title: Animate Your Thoughts: Decoupled Reconstruction of Dynamic Natural Vision from Slow Brain Activity Authors: Yizhuo Lu, Changde Du, Chong Wang, Xuanliu Zhu, Liuyun Jiang, Xujin Li, and Huiguang He Prior Knowledge Brain decoding with fMRI: Functional MRI measures brain activity through BOLD (Blood-Oxygen-Level Dependent) signals, which reflect neural activation across 3D voxel grids. Decoding methods aim to map these signals back to perceptual experiences like images or videos. Contrastive learning and CLIP: Contrastive loss functions (e.g., InfoNCE) train models to align representations from different modalities. CLIP is a pre-trained vision-language model that embeds text and images into a shared space, useful for representing high-level semantics. VQ-VAE (Vector Quantized Variational Autoencoder): A generative model that discretizes image data into latent tokens, enabling compression and reconstruction. It’s commonly used in diffusion models for reducing the input dimensionality while preserving structure. Transformer and sparse causal attention: Transformers model long-range dependencies via attention. Sparse causal attention masks future tokens and restricts attention to a sparse set, ensuring temporal consistency while reducing computational cost—critical for decoding motion from fMRI. Stable Diffusion and U-Net: Stable Diffusion is a text-to-image model that generates visuals via iterative denoising in a latent space. Its U-Net backbone processes image latents at multiple resolutions and is commonly inflated for video generation. Goal To accurately reconstruct dynamic natural vision (videos) from fMRI recordings by explicitly disentangling and decoding semantic, structural, and motion features — each aligned with distinct neural processes — in order to improve both reconstruction fidelity and neurobiological interpretability. ...

[This review is intended solely for my personal learning] Paper Info arXiv: 2308.10248 Title: Steering Language Models With Activation Engineering Authors: Alexander Matt Turner, David Udell, Nantas Nardelli, Sam Ringer, Tom McGrath, Eric Michaud, Mantas Mazeika Prior Knowledge Steering Language Models: Traditionally achieved through techniques such as prompt engineering, fine-tuning, and reinforcement learning from human feedback (RLHF), which often require substantial computational resources and specialized datasets. Internal Activation Manipulation: Exploring internal activations of neural networks can offer fine-grained control over their outputs without the cost associated with retraining or extensive model tuning. Goal The paper aims to introduce and validate a lightweight inference-time technique, termed Activation Addition (ActAdd), for steering the output of large language models (LLMs). This method targets latent capabilities of LLMs, such as controlled sentiment expression or reduced output toxicity, without the need for additional training. ...

[This review is intended solely for my personal learning] Paper Info arXiv: 2310.01405v4 Title: Representation Engineering: A Top-Down Approach to AI Transparency Authors: Evan Hubinger, Ajeya Cotra, Buck Shlegeris, Tom Lieberum, Nicholas Joseph, Owain Evans, Nicholas Schiefer, Oliver Zhang, Jan Brauner, Collin Burns, Leo Gao, Ryan Greenblatt Prior Knowledge Interpretability in AI: Traditional approaches often focus on low-level features such as neurons or circuits. However, such bottom-up analysis struggles with high-level abstractions like honesty or power-seeking. Representation Learning: Neural networks form internal embeddings of concepts during training, enabling powerful generalization and emergent behavior. Mechanistic Interpretability: Aims to reverse-engineer model internals to understand reasoning processes, but often lacks tools for directly modifying internal representations to improve safety. Transparency and Alignment: Transparency is critical for ensuring model behavior is aligned with human intentions and values, particularly to avoid deceptive alignment and inner misalignment. Goal The paper proposes and develops Representation Engineering (RepE), a new top-down framework for interpretability and model control. Rather than focusing on neurons or circuits, RepE centers on representations of high-level cognitive concepts and aims to understand, detect, and manipulate them. The core goal is to advance transparency methods that are directly useful for improving AI safety, including controlling undesirable behaviors like deception or power-seeking. ...

[This review is intended solely for my personal learning] Paper Info DOI: 10.1007/978-3-031-05409-9_23 Title: A Swarm Intelligence Approach: Combination of Different EEG-Channel Optimization Techniques to Enhance Emotion Recognition Authors: Sabahudin Balic, Lukas Kleybolte, and Christian Märtin Prior Knowledge EEG-based Emotion Recognition: EEG captures electrical brain activity via multiple electrodes. Its high temporal resolution makes it suitable for decoding emotional states in real-time. However, the full 32-channel setup is often computationally expensive and redundant. Swarm Intelligence: Algorithms like Particle Swarm Optimization (PSO), Cuckoo Search (CS), and Grey Wolf Optimizer (GWO) mimic social behavior of animals to solve optimization problems. Feature vs. Channel Selection: Traditional feature selection targets discriminative features across frequency bands, while channel selection focuses on spatially optimizing electrode positions. Goal To evaluate and compare different EEG channel selection techniques—both classical and swarm intelligence-based—to optimize emotion classification performance while significantly reducing computation time. ...

[This review is intended solely for my personal learning] Paper Info DOI: 10.1101/2024.11.17.623979 Title: Predicting whole-brain neural dynamics from prefrontal cortex fNIRS signal during movie-watching Authors: Shan Gao, Ryleigh Nash, Shannon Burns, Yuan Chang Leong Prior Knowledge Functional near-infrared spectroscopy (fNIRS) offers a more accessible alternative to functional magnetic resonance imaging (fMRI) but is limited by shallow cortical penetration. Prior research has demonstrated potential in predicting deep-brain fMRI signals from fNIRS data through linear predictive models trained on simultaneous fMRI and fNIRS measurements. ...