Analysis of learned representations has a blind spot: it focuses on $similarity$, measuring how closely embeddings align with external references, but similarity reveals only what is represented, not whether that structure is robust. We introduce $geometric$ $stability$, a distinct dimension that quantifies how reliably representational geometry holds under perturbation, and present $Shesha$, a framework for measuring it. Across 2,463 configurations in seven domains, we show that stability and similarity are empirically uncorrelated ($ρ\approx 0.01$) and mechanistically distinct: similarity metrics collapse after removing the top principal components, while stability retains sensitivity to fine-grained manifold structure. This distinction yields actionable insights: for safety monitoring, stability acts as a functional geometric canary, detecting structural drift nearly 2$\times$ more sensitively than CKA while filtering out the non-functional noise that triggers false alarms in rigid distance metrics; for controllability, supervised stability predicts linear steerability ($ρ= 0.89$-$0.96$); for model selection, stability dissociates from transferability, revealing a geometric tax that transfer optimization incurs. Beyond machine learning, stability predicts CRISPR perturbation coherence and neural-behavioral coupling. By quantifying $how$ $reliably$ systems maintain structure, geometric stability provides a necessary complement to similarity for auditing representations across biological and computational systems.

"An idea is nothing more nor less than a new combination of old elements" (Young, J.W.). The widespread adoption of Large Language Models (LLMs) and publicly available ChatGPT have marked a significant turning point in the integration of Artificial Intelligence (AI) into people's everyday lives. This study explores the capability of LLMs in generating novel research ideas based on information from research papers. We conduct a thorough examination of 4 LLMs in five domains (e.g., Chemistry, Computer, Economics, Medical, and Physics). We found that the future research ideas generated by Claude-2 and GPT-4 are more aligned with the author's perspective than GPT-3.5 and Gemini. We also found that Claude-2 generates more diverse future research ideas than GPT-4, GPT-3.5, and Gemini 1.0. We further performed a human evaluation of the novelty, relevancy, and feasibility of the generated future research ideas. This investigation offers insights into the evolving role of LLMs in idea generation, highlighting both its capability and limitations. Our work contributes to the ongoing efforts in evaluating and utilizing language models for generating future research ideas. We make our datasets and codes publicly available.

Graph autoencoders (GAEs), as a kind of generative self-supervised learning approach, have shown great potential in recent years. GAEs typically rely on distance-based criteria, such as mean-square-error (MSE), to reconstruct the input graph. However, relying solely on a single reconstruction criterion may lead to a loss of distinctiveness in the reconstructed graph, causing nodes to collapse into similar representations and resulting in sub-optimal performance. To address this issue, we have developed a simple yet effective strategy to preserve the necessary distinctness in the reconstructed graph. Inspired by the knowledge distillation technique, we found that the dual encoder-decoder architecture of GAEs can be viewed as a teacher-student relationship. Therefore, we propose transferring the knowledge of distinctness from the raw graph to the reconstructed graph, achieved through a simple KL constraint. Specifically, we compute pairwise node similarity scores in the raw graph and reconstructed graph. During the training process, the KL constraint is optimized alongside the reconstruction criterion. We conducted extensive experiments across three types of graph tasks, demonstrating the effectiveness and generality of our strategy. This indicates that the proposed approach can be employed as a plug-and-play method to avoid vague reconstructions and enhance overall performance.

The Connectivity Map (CMap) is a large publicly available database of cellular transcriptomic responses to chemical and genetic perturbations built using a standardized acquisition protocol known as the L1000 technique. Databases such as CMap provide an exciting opportunity to enrich drug discovery efforts, providing a 'known' phenotypic landscape to explore and enabling the development of state of the art techniques for enhanced information extraction and better informed decisions. Whilst multiple methods for measuring phenotypic similarity and interrogating profiles have been developed, the field is severely lacking standardized benchmarks using appropriate data splitting for training and unbiased evaluation of machine learning methods. To address this, we have developed 'Leak Proof CMap' and exemplified its application to a set of common transcriptomic and generic phenotypic similarity methods along with an exemplar triplet loss-based method. Benchmarking in three critical performance areas (compactness, distinctness, and uniqueness) is conducted using carefully crafted data splits ensuring no similar cell lines or treatments with shared or closely matching responses or mechanisms of action are present in training, validation, or test sets. This enables testing of models with unseen samples akin to exploring treatments with novel modes of action in novel patient derived cell lines. With a carefully crafted benchmark and data splitting regime in place, the tooling now exists to create performant phenotypic similarity methods for use in personalized medicine (novel cell lines) and to better augment high throughput phenotypic screening technologies with the L1000 transcriptomic technology.

We present a multiway fusion algorithm capable of directly processing uncertain pairwise affinities. In contrast to existing works that require initial pairwise associations, our MIXER algorithm improves accuracy by leveraging the additional information provided by pairwise affinities. Our main contribution is a multiway fusion formulation that is particularly suited to processing non-binary affinities and a novel continuous relaxation whose solutions are guaranteed to be binary, thus avoiding the typical, but potentially problematic, solution binarization steps that may cause infeasibility. A crucial insight of our formulation is that it allows for three modes of association, ranging from non-match, undecided, and match. Exploiting this insight allows fusion to be delayed for some data pairs until more information is available, which is an effective feature for fusion of data with multiple attributes/information sources. We evaluate MIXER on typical synthetic data and benchmark datasets and show increased accuracy against the state of the art in multiway matching, especially in noisy regimes with low observation redundancy. Additionally, we collect RGB data of cars in a parking lot to demonstrate MIXER's ability to fuse data having multiple attributes (color, visual appearance, and bounding box). On this challenging dataset, MIXER achieves 74% F1 accuracy and is 49x faster than the next best algorithm, which has 42% accuracy. Open source code is available at https://github.com/mit-acl/mixer.

Keypoint detection and description play a central role in computer vision. Most existing methods are in the form of scene-level prediction, without returning the object classes of different keypoints. In this paper, we propose the object-centric formulation, which, beyond the conventional setting, requires further identifying which object each interest point belongs to. With such fine-grained information, our framework enables more downstream potentials, such as object-level matching and pose estimation in a clustered environment. To get around the difficulty of label collection in the real world, we develop a sim2real contrastive learning mechanism that can generalize the model trained in simulation to real-world applications. The novelties of our training method are three-fold: (i) we integrate the uncertainty into the learning framework to improve feature description of hard cases, e.g., less-textured or symmetric patches; (ii) we decouple the object descriptor into two output branches -- intra-object salience and inter-object distinctness, resulting in a better pixel-wise description; (iii) we enforce cross-view semantic consistency for enhanced robustness in representation learning. Comprehensive experiments on image matching and 6D pose estimation verify the encouraging generalization ability of our method from simulation to reality. Particularly for 6D pose estimation, our method significantly outperforms typical unsupervised/sim2real methods, achieving a closer gap with the fully supervised counterpart. Additional results and videos can be found at https://zhongcl-thu.github.io/rock/

All of a sudden, everybody is talking about them irrespective of whether people are understanding the dissimilarity or not. Whether you are a follower of data driven science or not- you probably would have heard these terms. Let's start discussing and differentiating with a classic example of cats vs. dogs. In the picture below, can you identify cat and dog? Over the time, you have already seen many cats and dogs; and so, you are capable of identifying them as well. Also, we can not deny the fact that sometimes even humans make mistakes.