Overview:

We train a tiny ReLU network to output sparse top- distributions over a vocabulary much larger than its residual dimension. The trained network seems to converge to a mechanism closely resembling a Bloom filter: tokens are assigned sparse binary hashes, the hidden layer computes an approximate union indicator, and the output logits are linearly read from this union.

Here's what a small network trained on a toy version of the sparse top- distribution task learns to use:

Weight matrix of a 1-layer ReLU network trained via gradient descent on the toy -sparse distribution task below, for , , . Truncated at first tokens for visualisation purposes.

Plot of the range of values of , it forms a bimodal distribution.

That's the input weight matrix of the trained network. Every entry is either or . The network has effectively encoded a binary hash for each token - and as we'll show, this seems to enable the network to approximately simulate a Bloom filter, and so output the correct set of top- tokens with high probability.

We provide a theoretical construction showing how to set the weights to exactly implement a Bloom filter. The real network [...]

---

Outline:

(00:10) Overview:

(02:02) The Task:

(03:27) Construction:

(04:17) Formal construction:

(04:47) Analysis of a single forward pass:

(06:13) Training:

(07:04) Behavioural analysis of the trained network:

(10:14) Mechanistic analysis of the trained network:

(16:21) Conclusion / Reflections:

(18:24) Related work:

(19:25) Further work:

---

First published:

Source:

---

Narrated by TYPE III AUDIO.

---

Images from the article:

Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.